Trading: Traditional Theory, Modern Practice, and the SOTA Frontier

A self-contained learning report for a thoughtful generalist. Last revised May 2026.

Executive summary

Trading — the buying and selling of financial claims to make money — looks like one activity but actually contains several disjoint cultures held together by a shared price tape. A long-only mutual fund that holds Apple for ten years, a Citadel Securities market-maker quoting tens of millions of options contracts per day, a Yale endowment locking up capital in venture funds for fifteen years, a Renaissance Technologies statistical-arbitrage book turning over thousands of stocks daily, and a Telegram-coordinated meme-coin sniper running a MEV bot on Solana are all "trading." They share three problems — what to buy, when to buy it, and how to size the bet — and almost nothing else.

This report is a guided tour through all of it. It is organized so that a curious reader without quantitative training can walk from the foundational academic theory (Markowitz, Sharpe, Black-Scholes, Fama) through the modern practitioner toolkit (Avellaneda-Stoikov market-making, Almgren-Chriss execution, Lopez de Prado machine-learning workflows) into the contemporary frontier (transformer-based signals, autoencoder factor models, deep hedging, on-chain MEV, perpetual futures funding mechanics, the 0DTE options explosion). It uses algebra in the body and tells you exactly what to study next when the math goes beyond algebra.

A few orienting points that the rest of the report will defend in detail:

1. The "is the market efficient?" debate is the wrong question. The right question is the one Grossman and Stiglitz framed in 1980: markets are exactly as inefficient as they need to be to compensate the people who do the work of making them efficient. Almost every modern strategy lives in some corner of that gap. (Grossman & Stiglitz 1980)
2. Most published "factors" do not replicate. Harvey, Liu and Zhu cataloged 316 factors and argued that the conventional t-statistic threshold of 2.0 should be raised to roughly 3.0 to control for multiple testing. (Harvey, Liu & Zhu 2016; NBER w20592)
3. Machine learning has not replaced asset pricing, it has changed what asset pricing looks like. Gu, Kelly and Xiu (2020) showed that tree ensembles and neural networks roughly double the out-of-sample Sharpe of traditional linear factor models — but most of the gain comes from non-linear interactions among the same canonical predictors (momentum, liquidity, volatility), not exotic alternative data. (Gu, Kelly & Xiu 2020 RFS)
4. The structure of US equity trading is essentially three private firms (Citadel Securities, Jane Street, Virtu) and an exchange industry that competes with them. Jane Street alone took roughly 10% of US equity volume in 2024 and net trading revenue above $20 billion. (Global Trading; Bloomberg)
5. Crypto is now a fully professional asset class with its own market microstructure. Perpetual swaps are the dominant instrument, funding rates are the dominant carry signal, and MEV is the on-chain analog of payment-for-order-flow. The total addressable derivative volume on Hyperliquid alone routinely exceeds the entire US single-stock options industry on busy days.
6. The fastest-growing options product in history is the zero-days-to-expiry (0DTE) SPX option. It now accounts for roughly half of S&P 500 index option volume, up from ~5% in 2016. Its market-making microstructure is a live research frontier. (Cboe Volatility Insights)

The report is long because the field is. Use the table of contents to navigate; each subsection is self-contained.

Research brief

• Topic. Trading practices and theories, both traditional and current state-of-the-art (2024–2026), across all asset classes (equities, fixed income, FX, commodities, options, futures, crypto), all participant types (retail through institutional), both discretionary and systematic styles, and including DeFi.
• Audience. A highly capable self-learner who knows the casual definitions of "stock," "bond," and "option" but has no quantitative or markets background. Math kept at the algebra level in the body, with explicit "study next" pointers when the underlying math is heavier.
• Length target. Encyclopedic — 20,000–40,000 words. Designed to be revisited section by section.
• Recency policy. Mixed. The foundational canon (Markowitz 1952, Sharpe 1964, Black-Scholes-Merton 1973, Fama EMH 1970, Fama-French 1993, Carhart 1997, Lopez de Prado AFML) is presented in full, and the 2021–2026 SOTA (machine learning, alt data, transformer-based signals, modern microstructure, MEV, DEX/AMM mechanics, perpetual protocols, current factor critiques) is woven through.
• Math policy. Algebra only in body. Black-Scholes is shown in final form with each variable in plain English; the derivation (which requires stochastic calculus, Itô's lemma, partial differential equations) is referenced as "study next." Factor regressions, VaR/ES, Sharpe, drawdowns, AMM impermanent loss, funding-rate computations are all done in algebra with worked numbers.
• Source policy. Primary academic finance (Journal of Finance, Review of Financial Studies, Journal of Financial Economics, NBER), SSRN preprints, regulator publications (SEC, CFTC, FINRA, BIS, Federal Reserve), practitioner research (AQR, Two Sigma, Bridgewater, Lopez de Prado), respected commentators (Matt Levine, FT Alphaville, Hoffstein), and for crypto: Flashbots, Paradigm, a16z, Vitalik's blog, primary exchange docs.
• Out of scope. Trading regulation in non-US jurisdictions beyond passing reference, individual stock recommendations, the politics of taxation, comprehensive coverage of macroeconomic policy (which lives in a separate report in this series).

Key findings

1. The efficient market hypothesis (EMH) is best understood as a baseline, not a verdict. Fama (1970) gave the three forms (weak, semi-strong, strong), Grossman-Stiglitz (1980) showed strong-form efficiency is logically impossible, and Lo's Adaptive Markets Hypothesis (2004, expanded 2017) reframes efficiency as a state-dependent property of evolving market ecologies. (Fama 1970; Lo 2004 JPM; Princeton Press)
2. The "factor zoo" is real. Of ~316 documented anomalies, only a minority survive multiple-testing corrections, out-of-sample replication, and post-publication trading. The Hou-Xue-Zhang q-factor model (2015) and Stambaugh-Yuan mispricing factors (2017) explain large fractions of the survivors. (Hou-Xue-Zhang 2015 RFS; Stambaugh-Yuan 2017 RFS)
3. Black-Scholes-Merton is wrong in detail but right in scaffolding. No serious option trader uses constant volatility, but the model defines the language (Greeks, implied volatility surface) every options market participant speaks. The volatility smile/skew is the empirical fingerprint of where Black-Scholes fails. (Wikipedia: BSM model)
4. Modern execution is an optimization problem, not a clerical task. Almgren-Chriss (2000) framed the trade-off between market impact (trading fast hurts the price you get) and timing risk (trading slowly exposes you to adverse moves). Almost every institutional execution algorithm today descends from that mean-variance framing. (Almgren-Chriss 2000)
5. Renaissance Technologies' Medallion Fund is the empirical refutation of strong-form EMH for at least one firm. What is publicly known: short holding periods, thousands of weak signals combined, p < 0.01 thresholds, ruthless statistical hygiene, and a refusal to grow the fund beyond what its strategies can absorb without alpha decay. The actual signals remain proprietary. (Quartr breakdown)
6. Machine learning works in trading mostly as a non-linear factor compressor. Gu, Kelly and Xiu (2020) found tree models and neural networks roughly double the Sharpe of leading linear models, primarily by capturing interactions among the same canonical predictors. Autoencoder asset pricing (Gu, Kelly, Xiu 2021) generalizes Fama-French to latent, characteristic-driven factors. (Gu-Kelly-Xiu 2020; Autoencoder paper)
7. Backtest overfitting is the most dangerous failure mode in quantitative trading. Bailey and Lopez de Prado's deflated Sharpe ratio (2014), purged/embargoed k-fold cross-validation, and combinatorial purged cross-validation are now the de-facto standards in serious quant shops. (Bailey-Lopez de Prado 2014)
8. LLMs can extract trading-relevant signal from text. Lopez-Lira and Tang (2023) showed GPT-class models predict the direction of next-day returns from news headlines well above chance, with adoption-driven alpha decay already visible. (Lopez-Lira & Tang 2023)
9. Crypto market structure has matured into something quantitatively literate practitioners can model. Constant-product AMMs (Uniswap V2 x·y = k), concentrated liquidity (Uniswap V3), perpetual swaps with funding rates (dYdX, Hyperliquid, GMX), and MEV (Flashbots, MEV-Boost, proposer-builder separation) all have well-understood mechanics that map cleanly onto traditional microstructure concepts. (Hyperliquid docs via 21Shares; Flashbots)
10. 0DTE SPX options now exceed 50% of S&P index option volume. The structural impact on intraday volatility is contested: Cboe analyses argue the flow is more balanced than critics claim, but dealer gamma positioning is now a daily preoccupation of macro traders. (Cboe; Numerix)
11. Expected Shortfall has replaced VaR as the regulatory market-risk standard. Basel's Fundamental Review of the Trading Book (FRTB), in force from January 2025 in the EU and UK, uses 97.5% ES rather than 99% VaR because ES is sub-additive and tail-sensitive. (BPI)

Detailed analysis

Part I — Theory foundations

1. Modern Portfolio Theory (Markowitz, 1952)

Before 1952, "portfolio" advice was qualitative: don't put all your eggs in one basket. Harry Markowitz's Portfolio Selection (Journal of Finance, 1952) gave the slogan an equation, and won him a share of the 1990 Nobel Prize in Economics. (Markowitz model overview)

The core insight: when you combine two assets that don't move in perfect lockstep, the portfolio's volatility is less than the average of the two volatilities. The mathematical reason is that variance is not additive across correlated assets — the covariance term subtracts when correlations are below 1.

Expected return of a portfolio of two assets:

E[R_p] = w₁ · E[R₁] + w₂ · E[R₂]

where w₁ and w₂ are the weights (fractions of the portfolio) in assets 1 and 2 and E[R_i] is the expected return of asset i. Weights sum to 1.

Variance of a portfolio of two assets:

σ²_p = w₁²σ₁² + w₂²σ₂² + 2·w₁·w₂·ρ₁₂·σ₁·σ₂

Here:

• σ_i is the standard deviation (volatility) of asset i.
• ρ₁₂ is the correlation between the two assets (a number between -1 and +1).
• The third term is the covariance term, the heart of diversification.

Worked example. Asset A has σ = 20% per year, asset B has σ = 20%. You hold 50/50. If ρ = 1 (perfect correlation), portfolio σ is 20%. If ρ = 0, portfolio σ = √(0.25·0.04 + 0.25·0.04 + 0) = √0.02 ≈ 14.1%. If ρ = -1, portfolio σ = 0%. You have eliminated risk by combining negatively correlated assets at equal weight, without lowering expected return. This is "the only free lunch in finance."

The efficient frontier. Plot every possible combination of risky assets on a graph with volatility on the x-axis and expected return on the y-axis. The upper-left envelope of that point cloud — the set of portfolios with the highest expected return for each level of volatility — is the efficient frontier. Adding a risk-free asset (T-bills) and drawing a tangent line from the risk-free rate to the frontier gives the capital allocation line (CAL); the tangency point is the optimal risky portfolio in mean-variance terms. (Tidy Finance Markowitz overview)

What MPT actually requires you to know. The full mean-variance optimization needs an expected-return vector for every asset, a covariance matrix for every pair, and a risk-aversion parameter. In practice, all three are estimated from historical data with massive error bars. The optimizer is famously sensitive to small changes in expected returns — a property known as "error maximization." Practitioners use Bayesian shrinkage (Black-Litterman, 1992), risk parity, or 1/N as defensive alternatives. (Stanford notes on mean-variance with unknown moments)

Study next: convex optimization, quadratic programming, shrinkage estimators (Ledoit-Wolf), Black-Litterman.

2. The Capital Asset Pricing Model (Sharpe 1964, Lintner 1965, Mossin 1966)

CAPM extends Markowitz from "how do I build a portfolio?" to "what should any asset's expected return be in equilibrium?" The answer is the most famous equation in finance after Black-Scholes:

E[R_i] = R_f + β_i · (E[R_m] − R_f)

Where:

• E[R_i] = expected return on asset i.
• R_f = risk-free rate (typically a short-dated government bond).
• E[R_m] = expected return on the market portfolio (in practice, a broad index like the S&P 500).
• (E[R_m] − R_f) = the equity risk premium; historically ~5% in US equities.
• β_i = the asset's beta: its sensitivity to market moves.

Beta in plain English. If a stock has β = 1.5, then on a day the market goes up 1%, the stock is expected to go up 1.5%; if the market falls 1%, the stock falls 1.5%. β = 0 means market-neutral; β < 0 means it moves opposite the market (rare; gold sometimes, long-dated Treasuries during equity crashes).

Algebraically computing beta:

β_i = Cov(R_i, R_m) / Var(R_m)

You estimate this by regressing the stock's excess returns on the market's excess returns and reading off the slope coefficient. The intercept of that regression is Jensen's alpha — the manager's excess return after correcting for market exposure.

The security market line (SML) plots β on the x-axis and E[R] on the y-axis; CAPM says every asset must lie on the line. Empirically, the SML is too flat: low-beta stocks earn more than CAPM predicts, high-beta stocks earn less. This is the famous "betting against beta" anomaly that Frazzini and Pedersen (2014) turned into a strategy.

CAPM's empirical track record. Roll's critique (1977) argued the model cannot be tested because the "true" market portfolio is unobservable (it should include human capital, real estate, art, private equity, etc., not just listed stocks). Fama-MacBeth (1973) cross-sectional tests showed beta does relate to average returns, but not exactly along the CAPM line. The size effect (Banz 1981), value effect (Stattman 1980, Rosenberg-Reid-Lanstein 1985), and momentum effect (Jegadeesh-Titman 1993) all violated CAPM. CAPM survives as a teaching tool and a back-of-envelope cost-of-capital estimator, not as a model of expected returns.

Study next: Fama-MacBeth cross-sectional regressions, conditional CAPM (Jagannathan-Wang 1996), consumption CAPM (Lucas 1978, Breeden 1979).

3. Efficient Market Hypothesis (Fama, 1970; revisited)

Fama's 1970 Journal of Finance survey, Efficient Capital Markets: A Review of Theory and Empirical Work, became the foundational reference for asking "do prices reflect information?" Fama defined three nested forms (Fama 1970 summary; Fama 1991 Efficient Capital Markets II):

• Weak-form efficiency: prices already reflect all information contained in past prices. Implication: technical analysis using past price data alone cannot consistently earn abnormal returns.
• Semi-strong-form efficiency: prices reflect all publicly available information — past prices plus earnings, accounting data, news. Implication: fundamental analysis based on public data cannot earn abnormal returns.
• Strong-form efficiency: prices reflect all information, public and private. Implication: not even insiders can earn abnormal returns.

The joint hypothesis problem. Fama himself emphasized that you can never test market efficiency alone. You always test "efficiency and a particular model of expected returns." If you find that small stocks beat CAPM expectations, you cannot tell whether the market is inefficient or whether CAPM is the wrong asset-pricing model.

Anomalies that broke (or at least bent) EMH. The post-earnings-announcement drift (Bernard-Thomas 1989), the momentum effect (Jegadeesh-Titman 1993), the long-run reversals of De Bondt-Thaler (1985), the value premium (Fama-French 1992), the closed-end fund discount, the equity premium puzzle (Mehra-Prescott 1985), and the volatility puzzle (Shiller 1981) all challenged the joint hypothesis of efficient markets plus rational expected-return models.

Grossman-Stiglitz (1980): the impossibility of perfect efficiency. In a 1980 American Economic Review paper, Sanford Grossman and Joseph Stiglitz pointed out a logical paradox: if prices perfectly reflect all information, no one is paid for collecting information, so no one collects it, so prices reflect nothing. There must be an equilibrium degree of disequilibrium in which gathering information is just barely profitable enough to keep someone doing it. (Grossman-Stiglitz 1980; Wikipedia: Grossman-Stiglitz paradox)

This is the single most useful frame for understanding modern markets. Renaissance Technologies, Citadel, Two Sigma, and every other systematic shop is paid to do the work of making prices efficient — and they earn only as long as enough other participants do not.

The Adaptive Markets Hypothesis (Lo, 2004; book 2017). Andrew Lo's reframing treats market efficiency as an evolutionary property. Markets are populated by "species" (retail investors, pension funds, hedge funds, market makers, central banks) using heuristics that work in some environments and fail in others. Efficiency varies with the environment; bubbles and crashes are episodes when a regime change outpaces the speed of adaptation. (Lo 2004 JPM; Princeton: Adaptive Markets; Wikipedia: AMH)

The practical takeaway: an alpha that works for ten years can stop working when the population of competing strategies changes (think of statistical arbitrage in equities returning lower Sharpe in the late 2000s as more capital crowded the trade).

Study next: Shiller's Irrational Exuberance; Thaler's Misbehaving; behavioral finance (prospect theory, Kahneman-Tversky 1979); noise-trader models (De Long, Shleifer, Summers, Waldmann 1990).

4. Factor models: from Fama-French to the zoo

CAPM's failures pushed researchers to add factors that capture systematic risks beyond the market. The result was an arms race.

Fama-French 3-factor (1993). Eugene Fama and Kenneth French observed that small-cap stocks earn higher returns than large-caps, and high book-to-market ("value") stocks earn higher returns than low book-to-market ("growth") stocks, even after controlling for market beta. They constructed two zero-cost "factor" portfolios:

• SMB ("Small Minus Big"): long small-caps, short large-caps.
• HML ("High Minus Low"): long value, short growth.

The regression specification:

R_i − R_f = α + β_MKT · (R_m − R_f) + β_SMB · SMB + β_HML · HML + ε

The model explains around 90% of diversified-portfolio variance vs. CAPM's ~70% (in-sample, US data 1963–1991). (Wikipedia: Fama-French 3-factor)

Carhart (1997) 4-factor. Mark Carhart added UMD ("Up Minus Down"), a momentum factor: long stocks that rose most over the past year, short stocks that fell most. Persistence in mutual fund performance disappears once you control for momentum exposure — most "hot hands" are just funds that happened to ride a momentum theme. (Carhart 1997)

Fama-French (2015) 5-factor. Added two factors derived from profitability and investment:

• RMW ("Robust Minus Weak"): long high-profitability firms, short weak-profitability firms (~4.7% per year historically).
• CMA ("Conservative Minus Aggressive"): long low-investment firms, short high-investment firms (~3% per year). (Five-factor model SSRN)

The 5-factor model subsumes the value (HML) premium in many tests, making HML "redundant." Some practitioners now use a 6-factor model = 5-factor + momentum.

Hou-Xue-Zhang q-factor (2015). Kewei Hou, Chen Xue, and Lu Zhang built a 4-factor model grounded in investment-based asset pricing (q-theory):

• Market.
• Size.
• Investment (I/A).
• Return on equity (ROE).

The q-factor model explains roughly half of the ~80 anomalies it was tested against and outperforms Fama-French in many head-to-head comparisons. (Hou-Xue-Zhang 2015 RFS)

Stambaugh-Yuan (2017) mispricing factors. Robert Stambaugh and Yu Yuan grouped 11 well-known anomalies into two clusters — MGMT (management-driven: net issuance, accruals, etc.) and PERF (performance-driven: distress, momentum, profitability, etc.) — and showed the resulting 4-factor model (market, size, MGMT, PERF) accommodates more anomalies than Fama-French 5-factor or q-factor. (Stambaugh-Yuan 2017 RFS)

Asness-Frazzini-Pedersen (2019) Quality Minus Junk. AQR's three principals defined "quality" by a combination of safety (low leverage, low earnings volatility), profitability, growth, and payout. The QMJ factor — long high-quality stocks, short low-quality ("junk") — earns significant risk-adjusted returns in 24 countries. AQR's "quality at a reasonable price" funds package QMJ + value. (AFP 2019 Quality Minus Junk)

The "factor zoo" critique (Cochrane 2011; Harvey, Liu & Zhu 2016). John Cochrane's 2011 AFA presidential address coined "factor zoo" to describe the proliferation of anomalies. Harvey, Liu and Zhu (2016) catalogued 316 published factors and argued that with so much multiple testing, a t-statistic threshold of 2.0 is far too lenient — they propose a 3.0 minimum to control for the data-mining bias. They estimate that the majority of published factors are likely false discoveries. (Harvey-Liu-Zhu 2016; NBER w20592)

Hou, Xue and Zhang followed up in 2020 with a massive replication study finding that ~65% of anomalies fail to replicate at conventional significance after micro-cap and equal-weight corrections. Bryan Kelly and colleagues' machine-learning factor work (covered later) is partly an attempt to discipline this zoo with regularization rather than t-stat hurdles.

Practical implication. Most published anomalies are too small, too crowded, or too sensitive to micro-cap noise to trade profitably after costs. The premium factors that have survived replication, post-publication, and out-of-sample testing in multiple markets are essentially: market, size (weakly), value (cyclical), momentum, quality/profitability, low volatility, and investment. These are the "Fama-French-Carhart" workhorses that AQR, Dimensional, Avantis and others package as factor ETFs.

Study next: Cochrane's Asset Pricing (textbook), Harvey-Liu-Zhu's online updates, Bryan Kelly's machine-learning factor papers, Andrew Chen's "is everything a lie?" replication work.

5. Options pricing: Black-Scholes-Merton (1973)

In 1973, Fischer Black, Myron Scholes (Journal of Political Economy) and Robert Merton (Bell Journal) gave the world the closed-form pricing formula for European-style call and put options. Scholes and Merton won the 1997 Nobel; Black died in 1995, two years short. The formula transformed options from arcane instruments into a $700+ trillion notional global market.

The Black-Scholes call option formula:

C = S · N(d₁) − K · e^(−rT) · N(d₂)

with

d₁ = [ ln(S/K) + (r + σ²/2) · T ] / (σ · √T)
d₂ = d₁ − σ · √T

Each variable in plain English:

• C — the fair price today of a European call option (the right, not obligation, to buy one share for K at time T).
• S — current spot price of the underlying stock.
• K — strike price (the price the option holder can pay to receive the stock at expiry).
• r — annualized continuously compounded risk-free interest rate.
• T — time to expiry in years.
• σ — annualized volatility of the underlying's log returns.
• N(·) — cumulative standard normal distribution function (probability that a standard normal random variable is less than the argument).
• e^(−rT) — present-value discount factor.
• d₁ and d₂ are intermediate variables capturing how "in the money" and "in the money risk-adjusted" the option is.

Plain-English reading of the formula. The call's value equals the expected stock value at expiry given the option ends in the money (S · N(d₁)) minus the present-valued cost of paying the strike, weighted by the probability of exercise (K · e^(−rT) · N(d₂)). The two normal cumulative terms come from the assumption that log returns are normally distributed.

The put price (via put-call parity):

P = K · e^(−rT) − S + C

or equivalently,

P = K · e^(−rT) · N(−d₂) − S · N(−d₁)

Put-call parity itself, which holds independent of any pricing model under no-arbitrage and no-dividend assumptions, is:

C − P = S − K · e^(−rT)

Worked example. Stock at S = $100, strike K = $100, T = 0.25 years (3 months), r = 5% annual, σ = 20% annual.

• σ√T = 0.20 · 0.5 = 0.10
• d₁ = [ln(1) + (0.05 + 0.02)·0.25] / 0.10 = 0.0175 / 0.10 = 0.175
• d₂ = 0.175 − 0.10 = 0.075
• N(0.175) ≈ 0.5694
• N(0.075) ≈ 0.5299
• C ≈ 100·0.5694 − 100·e^(−0.0125)·0.5299 = 56.94 − 98.76·0.5299 ≈ 56.94 − 52.32 = $4.62.

A 3-month at-the-money call on a non-dividend stock with 20% vol is worth ~4.6% of spot.

Study next. The derivation requires stochastic calculus (Itô's lemma), the Black-Scholes partial differential equation, and a change of measure (Girsanov's theorem) to risk-neutral probability. Read Shreve's Stochastic Calculus for Finance II for the formal version. The intuition is: by continuously rebalancing a portfolio of stock and bonds, you can perfectly replicate the option's payoff under the model assumptions, so by no-arbitrage the option must cost the same as the replicating portfolio.

Binomial tree intuition (Cox-Ross-Rubinstein 1979). A discrete-time precursor: assume in one period the stock either goes up to Su (multiplied by u > 1) or down to Sd (multiplied by d < 1). Build a portfolio of Δ shares and B bonds whose value equals the option's value in both states. Solve. Take the limit as period length goes to zero, and you recover Black-Scholes.

One-period worked binomial. Stock at $100. In one month, goes up to $110 (u=1.1) or down to $90 (d=0.9). Strike $100 call. r = 0.5% per month.

Call payoff: $10 if up, $0 if down.

Risk-neutral probability: p = (e^(rΔt) − d) / (u − d) = (1.005 − 0.9) / (1.1 − 0.9) = 0.105/0.2 = 0.525.

Call value today: e^(−rΔt) · [p · 10 + (1−p) · 0] = 0.995 · 5.25 = $5.22.

Payoff diagrams. A long call has zero payoff below strike, then increases linearly with the underlying. A long put pays linearly as the underlying falls below strike, capped by the strike. Sold options have mirror-image payoffs (you collect premium upfront, then face open-ended losses).

Implied volatility (IV). The model takes σ as an input. Practitioners run it backwards: given the observed market price of an option, what σ must Black-Scholes use to match? That σ is the implied volatility. If Black-Scholes were correct, IV would be the same across strikes and maturities. It isn't.

Volatility smile/skew. Plot IV against strike for a given maturity. In equities, the curve typically slopes downward (deep out-of-the-money puts have higher IVs than at-the-money options) — the "volatility skew." Across currencies it's often more symmetric (a "smile"). The skew prices the empirical fat tail of returns and the demand for crash protection. The 1987 Black Monday crash is widely credited with creating the persistent equity skew. The whole topic of "stochastic volatility models" (Heston 1993, SABR, rough-volatility models like rBergomi) exists to fit the surface.

Vanna-Volga method. A practitioner technique to adjust Black-Scholes prices for the smile, especially in FX options. Roughly: hedge not just delta and gamma but also vega, vanna (sensitivity of delta to vol), and volga (sensitivity of vega to vol).

Study next: Hull's Options, Futures and Other Derivatives (textbook standard), Wilmott's Paul Wilmott on Quantitative Finance, Gatheral's The Volatility Surface, Bergomi's Stochastic Volatility Modeling. For derivations: Shreve Vol II; for measure theory: Williams' Probability with Martingales.

6. Fixed income mechanics

A bond is a contract to pay coupons periodically and return principal at maturity. The price of a bond is the present value of those cash flows under some discount rate. Trading bonds is largely the trading of interest rates and credit spreads.

Duration (Macaulay, 1938; modified duration is the practical version):

ModDuration ≈ −(1/P) · (dP/dy)

This measures the percentage change in price for a 1 percentage-point change in yield. A 10-year Treasury with modified duration ~8.5 means: if yields rise 1%, the price falls ~8.5%.

Convexity is the second-order correction:

ΔP/P ≈ −Duration · Δy + 0.5 · Convexity · (Δy)²

Convexity is always positive for vanilla bonds, meaning your loss when yields rise by 1% is less than your gain when yields fall by 1% — it's a benefit.

Yield curve construction. Build a continuous curve of zero-coupon yields by bootstrapping from quoted Treasury (or swap) prices at standard maturities. Modern desks use OIS curves (built from overnight index swaps) for discounting and SOFR/SONIA/€STR curves for floating-rate cash flows post-LIBOR-transition.

Credit spreads. A corporate bond yields more than a Treasury of the same maturity. The difference is the credit spread. There are several spread measures:

• Z-spread: the parallel shift you must add to the Treasury zero curve so that the discounted cash flows of the bond equal its market price.
• OAS (Option-Adjusted Spread): the Z-spread minus the value of any embedded options (callability for corporates, prepayment for mortgages). The right measure for bonds with optionality.

Other bond-trading concepts: roll-down (a bond gets shorter, the curve usually slopes up, so the bond's yield falls and its price rises if the curve stays still); butterfly trades (long 2yr and 10yr, short 5yr — a bet on curvature); basis trades (long the cheapest-to-deliver bond, short the Treasury futures contract — the "Treasury basis trade" was central to the March 2020 turmoil).

Study next: Bruce Tuckman & Angel Serrat's Fixed Income Securities: Tools for Today's Markets (the standard practitioner reference), Fabozzi's Handbook of Fixed Income Securities, the BIS papers on the Treasury market structure.

Mortgage-backed securities (MBS) and the prepayment puzzle. Roughly $9 trillion of US agency MBS — pools of mortgages guaranteed by Fannie Mae, Freddie Mac, or Ginnie Mae — trade as the second-largest fixed-income market after Treasuries. The wrinkle: homeowners have the right to refinance (a prepayment option). When rates fall, refinancing accelerates and MBS investors get their money back early at par, exactly when they would have most wanted to keep their high-coupon bonds. This negative convexity is the defining feature of MBS, and the reason MBS portfolio managers spend their careers building prepayment models. The CRT (credit-risk transfer) market, IO/PO strips, and TBA (to-be-announced) forwards together make MBS one of the richest single-asset-class ecosystems in finance.

The 2022–2024 rates rout. The most significant fixed-income event of the modern era was the Federal Reserve's 525bp hiking cycle from March 2022 through July 2023. The Bloomberg US Aggregate Bond Index lost ~13% in 2022 (its worst calendar year since the index began in 1976). 10-year Treasury yields rose from ~1.5% at end-2021 to >5% briefly in October 2023, the largest peacetime yield surge since the early 1980s. The episode wiped out Silicon Valley Bank (a textbook duration-mismatch failure), inflicted ~$685B of unrealized losses on US commercial banks at peak (FDIC data), and re-introduced bond traders born after 1985 to the meaning of duration risk.

7. Discretionary value investing (Graham-Dodd to Buffett to Greenblatt)

The intellectual ancestor of modern fundamental investing is Benjamin Graham, who with David Dodd published Security Analysis in 1934 and The Intelligent Investor in 1949. Three concepts dominate:

• Intrinsic value: the present value of the cash a business will throw off over its life. You estimate it by discounting projected cash flows. The estimate is always wrong; the question is how wrong and in which direction.
• Mr. Market: an emotionally unstable business partner offering daily quotes — sometimes too high, sometimes too low, never representing his "true" estimate of intrinsic value. Use him; don't trust him.
• Margin of safety: only buy when price is substantially below your intrinsic-value estimate, so that errors in estimation still leave you above water.

Warren Buffett's evolution. Early Buffett (1950s–1960s) was a strict Graham "cigar butt" investor — buying companies trading below liquidation value, regardless of business quality. His Berkshire Hathaway partnership letters from 1957–1969 are the canonical practical lectures.

Charlie Munger's influence pushed Buffett toward "wonderful companies at fair prices, not fair companies at wonderful prices." From the 1970s onward, the Berkshire approach became: buy compounders with durable competitive advantages ("moats"), pay reasonable prices, hold for very long periods. Examples: See's Candies, Coca-Cola, GEICO, BNSF Railway, Apple.

The arithmetic of Buffett's edge:

• Float from insurance (~$160B as of 2025) is essentially free leverage.
• Long holding periods minimize tax drag.
• Concentrated portfolio in deeply understood businesses.
• Behavioral discipline during crashes (the famous "I'm greedy when others are fearful" frame).

Joel Greenblatt's Magic Formula (in The Little Book that Beats the Market, 2006) is value investing reduced to two screens: rank stocks by earnings yield (EBIT/EV) and return on capital (EBIT/Tangible Capital), buy the top 30 by combined rank, hold for a year, repeat. Backtests on US equities 1988–2004 show ~30% annualized returns vs. ~12% for the S&P 500. Out-of-sample post-publication performance is more modest but still positive in most regimes.

The quality compounder thesis (Terry Smith, Nick Train, Chuck Akre): hold a small number of high-return-on-capital, low-debt businesses for decades and let compounding do the work. The thesis assumes the moat persists; the risk is multiple compression when growth slows.

Study next: Graham-Dodd Security Analysis (the 1940 second edition is the most practical), Buffett's annual letters (free at berkshirehathaway.com), Greenwald et al's Value Investing: From Graham to Buffett and Beyond, Damodaran's valuation textbooks.

8. Growth and momentum investing

The growth tradition emphasizes earnings trajectory rather than current cheapness.

• William O'Neil's CANSLIM (How to Make Money in Stocks, 1988): a checklist combining current quarterly earnings, annual earnings growth, new products/management, supply/demand (low float), leader vs. laggard, institutional sponsorship, and market direction. Quasi-systematic, with explicit chart-pattern entry triggers.
• Peter Lynch's "buy what you know" (One Up on Wall Street, 1989): scuttlebutt research on consumer-facing companies, with a strong bias for small-cap growth. Lynch ran Magellan to ~29% annualized 1977–1990, the best long-run mutual-fund record of the era.

The modern critique of growth investing. Returns from growth stocks have been concentrated in a small number of winners (Bessembinder 2018: only ~4% of US stocks accounted for all net wealth creation above T-bills since 1926). Most "growth" stories underperform — the survivorship bias in retrospective hagiographies of FAANG is severe.

9. Technical analysis — what works and what doesn't

Technical analysis (TA) is the family of methods that use only price and volume data, often with visual chart-pattern recognition. Its core claim is that historical price action contains predictive information beyond what fundamental analysis captures.

The honest scorecard.

• Time-series momentum — the simple rule "buy what's been going up over the past 1–12 months, sell what's been going down" — has the strongest empirical support of any TA-adjacent strategy. Moskowitz, Ooi and Pedersen (2012) documented it across 58 instruments in 25+ years of data with extremely consistent results. (Moskowitz-Ooi-Pedersen 2012)
• Cross-sectional momentum (Jegadeesh-Titman 1993) — rank stocks by past 6–12 month return, long the top decile, short the bottom decile — has also replicated robustly worldwide, with significant drawdowns ("momentum crashes" of Daniel-Moskowitz 2016).
• Moving-average crossover trend-following has positive expected returns in many markets, particularly futures, but is essentially a noisy implementation of time-series momentum. Faber's "The Quantitative Approach to Tactical Asset Allocation" (2007) is the most-cited retail-accessible version.
• Chart patterns (head-and-shoulders, double tops, flags, pennants): subjective by construction, with mixed evidence. Lo, Mamaysky and Wang (2000) used non-parametric kernel regression to formalize ten patterns and found some statistical content, but with effect sizes too small to support discretionary trading.
• RSI, MACD, Bollinger Bands: as standalone signals, none have strong out-of-sample evidence. Used as filters on top of trend-following strategies they can marginally improve risk-adjusted returns.
• Elliott Wave, Gann angles, Fibonacci retracements: no rigorous statistical support. Treat as folklore. (CFA Institute literature review on TA)

The honest summary: trend-following has rigorous empirical support, and most "chartism" does not. The CFA Institute literature review (2016) and the broader academic consensus agree.

10. Discretionary global macro

Global macro takes top-down views on currencies, rates, commodities, and equities, sized for conviction.

• George Soros's reflexivity (The Alchemy of Finance, 1987): markets and fundamentals influence each other. Rising prices change behavior (more investment, more borrowing, more confidence) that changes the fundamentals that prices were "supposed" to reflect. The 1992 ERM trade ($1B+ profit shorting sterling) is the canonical example.
• Stanley Druckenmiller: "the way to build wealth is to preserve your capital and wait for the few opportunities to make extraordinary returns." Concentrated, infrequent bets, ruthless cutting of losers, willingness to flip from long to short overnight.
• Ray Dalio's Principles and the "economic machine": a systematic frame for thinking about credit cycles, deleveragings, and the interaction of growth, inflation, and policy. Bridgewater's flagship Pure Alpha expresses this discretionarily; All Weather (covered next) expresses the risk-balanced version systematically.

Study next: Sebastian Mallaby's More Money Than God (history of hedge funds), Jack Schwager's Market Wizards series, Soros's Alchemy of Finance.

10b. Behavioral finance — the cognitive backdrop

No serious 21st-century practitioner can ignore the body of research that shows real traders are not the rational maximizers of textbook utility theory.

• Prospect Theory (Kahneman & Tversky 1979). Decisions under risk are made relative to a reference point. Losses hurt about twice as much as equivalent gains feel good (loss aversion). People over-weight small probabilities and under-weight large ones. This shows up in trader behavior as: cutting winners early, riding losers too long, paying too much for OTM puts and lottery-ticket calls.
• Overconfidence and the trading-cost penalty (Odean 1999; Barber-Odean 2000–2002). Retail traders who trade more, earn less. The headline finding from a 60,000-account discount-broker study: gross returns matched the market, but net of trading costs the most-active quintile underperformed by ~6% per year. Men trade more than women, and earn less.
• Disposition effect (Shefrin-Statman 1985). The tendency to sell winners and hold losers. Empirically robust across markets, with negative average consequences for tax-aware investors.
• Anchoring (Tversky-Kahneman 1974). Once a price is "fixed" in the trader's mind (a 52-week high, a purchase cost, a round number), subsequent judgments cluster around it.
• Herding and information cascades. Bikhchandani-Hirshleifer-Welch (1992). When private signals are noisy and public action is observable, rational individuals may rationally ignore their private signals and follow the herd, creating fragile equilibria that can flip on weak evidence.
• Limits to arbitrage (Shleifer-Vishny 1997). Rational arbitrageurs cannot always correct mispricing because of capital constraints, redemption risk, and noise-trader risk — the arbitrageur can be right and still be carried out before the trade converges. This is the modern reason behavioral biases survive: someone might exploit them, but not necessarily fast enough.

The practical implication for the working trader: the most reliable source of edge is often not better predictions but better discipline against your own biases. Pre-commitment via systematic rules, position sizing tied to objective signals, and explicit drawdown limits are the cheap defensive technologies.

11. Risk-parity, all-weather, and the endowment model

Risk parity allocates capital so that each asset class contributes equally to total portfolio risk, not equally to capital. Because bonds have lower volatility than stocks, a risk-parity portfolio holds more bonds (often levered) than a 60/40 portfolio.

Bridgewater's All Weather (Ray Dalio, mid-1990s): the canonical risk-parity portfolio, designed to perform in any combination of (rising/falling growth) × (rising/falling inflation). The classic illustrative allocation: 30% stocks, 40% long-term Treasuries, 15% intermediate Treasuries, 7.5% gold, 7.5% diversified commodities, sometimes levered up to match equity volatility. (Bridgewater: The All Weather Story; Optimized Portfolio)

The intuition: a 60/40 portfolio's volatility is ~90% driven by stocks. Risk parity equalizes that. The 2022 stocks-and-bonds drawdown (when both fell together due to inflation) was uniquely bad for risk parity, prompting much hand-wringing; 2024 was its best year since 2019 as correlations normalized.

State Street and Bridgewater launched the SPDR Bridgewater All Weather ETF (ALLW) in March 2025, putting institutional risk parity in a listed wrapper.

Yale endowment model (David Swensen, 1985–2021) allocates 50–70% of capital to illiquid alternatives (private equity, venture capital, hedge funds, real assets) on the bet that an "illiquidity premium" rewards long-duration capital. Yale's endowment compounded at ~13.7% annually under Swensen, growing from $1B to $31B+. (Quantified Strategies; Chronograph)

The model's critique: the illiquidity premium has compressed as more endowments crowd in; access to top-quartile PE/VC managers is the actual moat, not the asset class. (Rebalance: Diminishing Returns for University Endowments)

60/40 critique and revival. The 60% stocks / 40% bonds benchmark portfolio was declared dead repeatedly during 2010–2021 (low bond yields meant the 40% had no carry), and especially in 2022 (stocks and bonds fell together for the first sustained period since the 1970s). 2023–2025 saw 60/40 recover dramatically as bond yields normalized to historically attractive levels.

Part II — Asset-class practices

12. Equities

Cash equities. US equity markets clear through fragmented venues: 16 lit exchanges (NYSE, Nasdaq, IEX, Cboe BZX/BYX/EDGA/EDGX, MEMX, MIAX Pearl, LTSE, NYSE National/American/Arca/Chicago, Members Exchange, etc.), 30+ alternative trading systems (ATSs, mostly dark pools), and major wholesalers (Citadel Securities, Virtu, G1, Two Sigma Securities) that internalize most retail flow via payment-for-order-flow (PFOF) arrangements.

Order types.

• Market order — execute immediately at the best available price.
• Limit order — execute only at a specified price or better; rests in the order book until filled or canceled.
• Stop order — becomes a market order when a trigger price is hit.
• IOC/FOK — Immediate-Or-Cancel / Fill-Or-Kill discipline modifiers.
• Hidden / iceberg — display only a fraction of size.
• Midpoint peg — track the midpoint of the National Best Bid and Offer (NBBO).

The NBBO and Reg NMS. US equity markets are governed by Regulation NMS (2005), which created the National Best Bid and Offer and required brokers to route trades to the venue with the best displayed price (the "trade-through rule"). Reg NMS is the reason US equity microstructure looks like it does.

Where retail flow actually goes. When you click "buy 100 AAPL" in Robinhood, the order typically does not hit an exchange. It is sold to a wholesaler (most commonly Citadel Securities), which internalizes the trade — quoting a price slightly inside the NBBO, capturing the spread, and routing only flow it doesn't want to handle to the exchanges. PFOF — the wholesaler pays the broker for the order — is the dominant retail-broker revenue model in the US. The SEC's December 2022 proposal to ban PFOF was reduced in the final rule (the "tick size" and "minimum increment" pieces). (SEC Reg NMS amendments and related)

13. Equity options

The US listed options market trades on 17 exchanges and clears through the Options Clearing Corporation (OCC). Daily contract volume routinely exceeds 50 million contracts; SPX options alone trade over 2 million contracts/day, of which a majority are now 0DTE (zero-days-to-expiry).

The Greeks. Sensitivities of option price to underlying factors:

• Delta (Δ): ∂Price/∂Spot. The hedge ratio.
• Gamma (Γ): ∂²Price/∂Spot². How fast delta changes — high near the strike, near expiry.
• Theta (Θ): ∂Price/∂Time. Time decay; almost always negative for long options.
• Vega (V): ∂Price/∂σ. Sensitivity to volatility.
• Rho (ρ): ∂Price/∂r. Interest-rate sensitivity.

Dealer Greeks ("dealer positioning") have become a daily macro tape — when SPX dealers are short gamma, they must sell into declines and buy into rallies, mechanically amplifying intraday volatility; when long gamma, they damp it. The growth of 0DTE has made this a much sharper effect (see "0DTE explosion" below).

Volatility surface. A 3D plot of implied vol against strike and maturity. Traders speak in terms of:

• ATM (at-the-money) vol.
• Skew (the slope of vol across strikes).
• Term structure (the slope across maturities; the "vol curve" is normally upward-sloping but inverts in stress).
• Smile (the curvature of the strike-direction slice).

Common option strategies.

• Covered call: long stock, short call. Capped upside, partial downside protection from premium. Retail favorite; "buy-write" indexes (e.g., BXM) track this systematically.
• Cash-secured put: short put with cash reserved to buy stock if assigned. Synthetically equivalent to a covered call by put-call parity.
• Straddle: long call + long put at same strike. Pays off on large moves either direction. The "long vol" trade.
• Strangle: like a straddle but with OTM call and put. Cheaper, wider profit zone.
• Iron condor: short OTM put spread + short OTM call spread. Range-bound bet.
• Calendar spread: long longer-dated option, short shorter-dated option of the same strike. Bet on term-structure shifts.

Volatility trading professionalized. The CBOE VIX index, computed from a portfolio of SPX options, is the most-watched volatility measure. VIX futures (launched 2004) trade ~250k contracts per day. The structural property that VIX futures usually trade in contango (term structure upward sloping) gives rise to the long-running "short vol" trade (sell VX1 futures, roll down the curve). It blew up spectacularly in February 2018's "volmageddon," when XIV (the inverse short-vol ETN) lost ~96% of NAV in one day. The lesson: short-vol strategies collect small steady premia and pay back occasionally in catastrophic chunks.

The dispersion trade. Sell vol on the index, buy vol on a basket of component stocks. Profits when realized correlations fall below implied — i.e., when individual names move idiosyncratically while the index is calm. It's a popular vol-arb strategy at firms like Susquehanna and Optiver.

The vol risk premium (VRP). The persistent gap between implied volatility and subsequently realized volatility. Across markets and time, implied vol is on average ~3 percentage points higher than realized. This is the empirical fingerprint of an insurance premium — option buyers, on average, overpay for crash protection. The VRP is the structural source of return for short-vol strategies. (Cremers, Halling, Weinbaum 2015 JFE and related VRP literature.)

14. Fixed income trading in practice

Cash bonds are still largely a voice and request-for-quote (RFQ) market, though electronification has progressed rapidly. Key venues:

• US Treasuries: traded inter-dealer on BrokerTec, Dealerweb, Fenics UST. Retail and institutional clients use Bloomberg AllQ, MarketAxess and Tradeweb.
• Credit (IG and HY corporate bonds): MarketAxess dominates electronic flow with all-to-all "Open Trading."
• Repo and securities lending: a $4–5 trillion daily market that finances most leveraged bond positions.

Treasury basis trade. Hedge funds (mainly Citadel, Millennium, and a few specialists) finance long positions in the cash-cheapest-to-deliver Treasury and short the corresponding Treasury futures contract, capturing the slight pricing wedge between the two. The trade is highly levered (often 50:1+ via repo) and is now an estimated $700B–$1T position. The Fed has flagged it as a financial-stability concern; the March 2020 dislocation forced the central bank to backstop Treasury liquidity. (OFR brief on Treasury basis and Fed Liberty Street commentary)

Credit trading dynamics. Most corporate bonds trade infrequently; a typical IG bond may trade only a few times per week. Liquidity is concentrated in the on-the-run (most recent issue) of large issuers. ETFs (LQD for IG, HYG for HY) have become the dominant liquidity layer — many institutional managers use ETF flow to hedge cash credit exposures.

15. FX

The FX market is the largest (about $7.5 trillion in average daily turnover per the 2022 BIS Triennial survey) and remains primarily inter-dealer. The structure:

• Top-tier banks (JPM, Citi, Deutsche, UBS, HSBC, Goldman) make markets to their corporate and institutional clients on single-dealer platforms.
• Multi-dealer platforms (EBS, Refinitiv FXall, Bloomberg FXGO) provide aggregated liquidity.
• Non-bank market makers (XTX Markets, Citadel Securities, HC Tech, Jump) now provide a substantial share of inter-bank liquidity, particularly in G10 spot.
• Retail aggregators (OANDA, IG, FXCM) intermediate retail demand, often with house market-making.

FX carry trade. Borrow in low-yielding currencies (historically JPY and CHF, more recently EUR or CNY in some windows) and lend in high-yielding currencies (historically AUD, NZD, BRL, ZAR). Empirically positive Sharpe over decades, but with severe negative skew — carry trades suffer massive losses in "risk-off" episodes (2008, March 2020). The Koijen-Moskowitz-Pedersen-Vrugt "Carry" paper (2018) generalizes this to commodities, equities, and bonds. (KMPV 2018 JFE)

Uncovered Interest Parity (UIP) puzzle. The theory says high-yield currencies should depreciate on average to offset the yield advantage. Empirically they don't — that's why carry works. KMPV reject UIP in favor of varying-risk-premium models.

16. Commodities

Commodities trade primarily through futures markets (CME Group, ICE) and OTC swaps. Three broad submarkets:

• Energy (WTI/Brent crude, natural gas, gasoline, heating oil): the most active futures contracts in the world by notional.
• Metals (gold, silver, copper, aluminum): split between CME (NY) and LME (London, with its physical-delivery quirks; recall the March 2022 nickel short squeeze).
• Agriculture (corn, soybeans, wheat, sugar, coffee, cocoa, cattle, hogs): seasonal, weather-driven, with substantial commercial-hedger flow.

Commodity term structure.

• Contango: futures price > spot price. Negative roll yield for long-only investors.
• Backwardation: futures price < spot price. Positive roll yield.

The same long-only investor in the same physical commodity earns very different returns depending on term-structure regime — the long-running headline that "oil tripled but USO didn't" is explained by USO's contango bleed.

Commodity factor strategies: time-series momentum and carry both work in commodities, and the basket-level returns to "alternative risk premia" in commodities have been roughly comparable to those in equities (with much higher volatility).

17. Crypto (covered in depth later — see Part V)

Part III — Quantitative and systematic methods

18. Statistical arbitrage and pairs trading

Statistical arbitrage ("stat arb") is the practice of taking many small, market-neutral bets based on statistical relationships among securities. It was pioneered in the late 1980s at Morgan Stanley's APT group (Bamberger, then Tartaglia, then a young David Shaw) and matured at Renaissance Technologies, D.E. Shaw, and PDT Partners.

Pairs trading. The canonical example: two related stocks (say, Coke and Pepsi, or two utilities) that historically trade in a stable ratio. When the ratio diverges, short the temporarily expensive one and go long the cheap one; profit when the ratio reverts.

Cointegration. Two non-stationary time series are cointegrated if some linear combination of them is stationary. Engle-Granger (1987) developed the test that gives this its name. In trading, cointegration is the statistical justification for pairs trading: the spread between two cointegrated series will tend to mean-revert.

The empirical record: classic pairs-trading Sharpe ratios were ~1.5–2 in the 1980s–1990s, decayed substantially by the mid-2000s as more capital entered. Modern stat-arb uses thousands of factors and machine learning rather than simple pairs, but the conceptual ancestor is the same.

Basket arbitrage generalizes pairs: instead of one stock vs. one stock, you trade a portfolio of N longs vs. a portfolio of M shorts whose net exposure to common risk factors is approximately zero, betting that the residual mean-reverts.

19. Market microstructure

Microstructure is the study of how prices form at the highest resolution — order by order, microsecond by microsecond.

Limit order books (LOBs). A LOB is a queue of resting orders: at each price level, orders are matched price-time priority (first by best price, then by time priority within a price level). The best bid is the highest buy limit; the best ask the lowest sell limit; the difference is the spread.

Queue position. On highly active stocks, the limit at the best bid may be huge. A new order joining at the back of the queue may take seconds or minutes to fill — by which time the price may have moved. Queue position is a tradeable asset. Specialized HFT firms invest enormous effort in maintaining favorable queue positions.

Market impact. When you buy a large quantity, your buys consume liquidity and push the price up. Impact is roughly square-root in size for institutional orders ("square-root law of market impact," empirically supported in many studies by JP Morgan, Capital Fund Management, and others).

Glosten-Milgrom (1985) model of the spread. In a market with informed and uninformed traders, the market maker cannot tell which kind any individual order comes from. Each buy raises the maker's posterior belief that the asset's true value is higher; each sell does the opposite. The maker sets ask above the unconditional mean (to break even on the upside-revising buys) and bid below it (to break even on the downside-revising sells). The spread is the adverse-selection cost. (Glosten-Milgrom 1985; Wikipedia: Glosten-Milgrom)

Kyle (1985) model. Albert Kyle's Econometrica paper modeled a single risk-neutral informed trader (the "insider") submitting orders to a market populated by noise traders and competitive market makers. The key quantity is Kyle's lambda (λ): the slope of the price impact of order flow. λ measures market depth and the rate at which private information is incorporated into prices. (Kyle 1985)

In equilibrium, the insider trades slowly (to disguise his information among the noise traders), prices follow a martingale, and all the private information is incorporated by the end of trading. Kyle's λ has become the standard empirical metric for market depth — a low λ market is "deep" (large orders barely move price), a high λ market is "shallow."

20. Optimal execution: Almgren-Chriss (2000)

Suppose you must liquidate 1 million shares of a stock by end-of-day. Trading all 1M at the open will smash the price. Trading 50,000 shares per half-hour over 20 intervals reduces impact but exposes you to bad news arriving while you wait.

Almgren and Chriss (2000) formalized this as a mean-variance optimization: minimize a weighted combination of expected execution cost (which goes up if you trade fast) and variance of execution cost (which goes up if you trade slow). The optimal trajectory is an exponentially-decaying schedule; the more risk-averse the trader, the faster the schedule. (Almgren-Chriss 2000)

This is the intellectual foundation of nearly every modern execution algorithm:

• VWAP (Volume-Weighted Average Price): match the daily volume profile. Common benchmark.
• TWAP (Time-Weighted Average Price): trade evenly through the day. Simplest possible benchmark; rarely optimal.
• Implementation Shortfall (IS): minimize the difference between the price at decision time and the realized average fill — directly the Almgren-Chriss formulation.
• POV (Percent-of-Volume): trade at a fixed percentage of market volume.
• Adaptive arrival-price / liquidity-seeking algos: modern derivatives using ML to forecast short-term price and volume.

Study next: Cartea, Jaimungal, and Penalva's Algorithmic and High-Frequency Trading (the textbook standard); Robert Almgren's published lectures.

21. High-frequency trading: what it actually is

"HFT" lumps together many distinct businesses. Roughly:

1. Electronic market making — quoting two-sided markets and earning the spread. Citadel Securities, Virtu, Jane Street (in ETFs), Susquehanna, Optiver, IMC.
2. Latency arbitrage — exploiting fleeting price discrepancies between venues (e.g., the SPY ETF momentarily diverges from its NAV, or the futures lead the cash). Jump Trading, Tower Research, DRW, Hudson River Trading.
3. Statistical arbitrage at HFT timescales — short-horizon mean-reversion or factor arbitrage held for milliseconds to minutes. Renaissance's Medallion (longer horizons than pure HFT), PDT, and the systematic books at the big multi-strats.
4. Liquidity provision to ETF authorized-participant ecosystems — Jane Street is the global leader; the firm essentially is the connective tissue between ETF mechanics and underlying basket markets.

The infrastructure arms race. Microwave towers between Aurora (CME futures) and Carteret (NJ data centers) cut latency from ~16 ms (fiber) to ~8 ms (microwave). McKay Brothers, Anova, and Vigilant Global operate competing microwave networks. The Spread Networks fiber from Chicago to NJ (Lewis's Flash Boys opens with it) is now obsolete; current best latency is microwave, with prospective satellite-based offerings.

Flash Boys, IEX, and the speed-bump debate. Michael Lewis's 2014 book Flash Boys told the story of Brad Katsuyama, an RBC trader who built IEX to neutralize the speed advantages of HFT through a 350-microsecond delay (a coiled fiber-optic "speed bump") between order arrival and matching engine. The book made HFT a political issue; IEX has captured a few percent of US equity market share but never threatened the incumbents. The SEC has wrestled repeatedly with whether speed bumps are consistent with Reg NMS. (Flash Boys analysis; Traders Magazine on IEX)

The honest summary of HFT. Spreads on US equities are dramatically lower than they were in the mid-1990s (mostly thanks to decimalization in 2001, but HFT has helped). Retail execution quality is in many ways the best in history. HFT firms compete by both narrowing markets (good for liquidity takers) and by adverse-selecting slower participants (bad for slower liquidity providers). The cumulative result is hotly debated; the SEC's market-structure proposals of 2022–2024 reflect ongoing tension.

Payment for order flow (PFOF). Wholesalers like Citadel and Virtu pay brokers (Robinhood, Schwab, etc.) for the right to internalize retail orders. The wholesaler captures the spread, after providing the retail customer with "price improvement" over the NBBO. Critics argue PFOF creates conflicts of interest; defenders point to the demonstrably better execution quality of retail orders today. The SEC's 2024 minimum-tick rules tightened spreads, partially undercutting the PFOF business model.

22. Market making: theory in practice

The market maker's problem: quote two-sided prices, capture the spread on average, and survive adverse selection without accumulating dangerous inventory.

The Avellaneda-Stoikov (2008) framework is the modern theoretical workhorse. Suppose the mid-price S follows a Brownian motion. Orders arrive at exponentially-decreasing rates as the maker's quote moves away from mid. The maker chooses bid and ask quotes to maximize expected utility of terminal wealth, penalizing variance of inventory. The solution prescribes:

• A reservation price = mid − inventory · risk-aversion · σ² · time-to-horizon. This shifts the maker's center away from the public mid in the direction that would unload inventory.
• A half-spread that depends on risk aversion, vol, time horizon, and order-arrival intensity.

The optimal bid is (reservation price − half spread); optimal ask is (reservation price + half spread). (Avellaneda-Stoikov 2008; Hummingbot guide)

In plain terms: when you're long too much inventory, lower your bid and your ask (to attract sells, discourage buys). The framework is the workhorse for crypto market making (the open-source Hummingbot package uses an A-S variant) and for many equity/futures shops.

Study next: Cartea-Jaimungal-Penalva (above), Guéant's The Financial Mathematics of Market Liquidity. The derivation requires Hamilton-Jacobi-Bellman PDEs.

23. Trend-following and time-series momentum

The trend-following industry — AHL, Man Group, Winton, Aspect, Millburn, Campbell, Lynx, Graham Global, MAN AHL — manages roughly $300B in CTA (commodity trading advisor) assets. The core strategy is "buy what has been going up, sell what has been going down" across diversified futures markets, scaled to a target volatility.

Moskowitz-Ooi-Pedersen (2012) documented the academic version: across 58 futures contracts over 25+ years, the past 12-month return positively predicts the next month's return. A portfolio that goes long instruments with positive past-year returns and short those with negative past-year returns, sized for volatility, generates ~1+ Sharpe with relatively low correlation to traditional asset classes. (Moskowitz-Ooi-Pedersen 2012 JFE)

The 2008 crisis was the trend-following industry's high-water mark — long bonds, short equities, short oil, all worked simultaneously. The 2010s were largely a drought for CTAs. 2022 was a magnificent year (long commodities, short bonds, short equities all worked), and trend-followers have outperformed since.

The "crisis alpha" claim. Trend-followers tend to perform well in extended directional crises (where many markets move persistently in one direction) but suffer in choppy, range-bound markets and at sharp turning points (Q1 2009, Q1 2016).

24. Carry as a cross-asset strategy

Carry = the return you earn if nothing changes. Examples:

• FX: hold a high-yield currency funded by a low-yield one; earn the interest differential.
• Bonds: hold longer-duration bonds when the yield curve is upward-sloping; earn roll-down.
• Equity: dividend yield + buyback yield.
• Commodity: backwardation — front-month futures cheaper than spot, so rolling forward earns the spread.
• Volatility: short vol earns the vol risk premium when realized vol < implied vol.

The KMPV (2018) "Carry" paper showed that, across asset classes, a simple "buy high carry / sell low carry" portfolio earns persistent positive returns with similar risk characteristics: positive expected returns, negative skew (crashes), exposure to global liquidity and volatility risks. (KMPV 2018)

The honest summary: carry is real and persistent, but you are being paid to bear tail risk. In a benign environment carry harvests pennies in front of a steam-roller; in crises it can give back years of gains in weeks.

25. Volatility trading

A full taxonomy:

• Short vol structures: short straddles/strangles, sold variance swaps, sold VIX futures, covered calls. Positive carry, negative skew.
• Long vol structures: long straddles, long OTM puts, long VIX futures, long-vol overlays. Negative carry, positive skew. Spitznagel/Universa's "Black Swan" tail-hedge fund is the most famous example. (Wikipedia: Universa; Yahoo Finance: Spitznagel)
• Dispersion: short index vol, long single-name vol — exposed to realized correlation falling below implied.
• Calendar / vol-of-vol: bet on the shape of the term structure or the volatility of volatility itself.
• Vol arbitrage: trade differences across markets or instruments (e.g., listed vs. OTC, single-name vs. ETF).

The empirical fingerprint of the vol risk premium (implied > realized by a few vol points on average) is one of the most robust findings in finance. The associated trade is also one of the most prone to blowup. The February 2018 "volmageddon" episode (XIV's terminal day, several -97% NAV moves in single-day inverse-vol ETPs) is the canonical case study.

26. Risk premia and "alternative risk premia"

Beginning in the early 2010s, banks (Goldman, JPM, Morgan Stanley, Société Générale) and asset managers (AQR, Man, BlackRock) wrapped academic factors — value, momentum, carry, low vol, quality, trend — into systematic indices and sold them as "Alternative Risk Premia" (ARP). The pitch: cheap, transparent, hedge-fund-like returns without 2-and-20 fees.

The 2018–2020 ARP "quant winter" was brutal: value cratered, momentum was choppy, low vol underperformed, and trend was middling. AUM in ARP products fell substantially. The 2022–2024 recovery has been quieter, partly because much of the institutional appetite moved on. The intellectual lesson: factor crowding is real; harvested premia decay; and "alternative" returns turn out to be highly correlated to standard factors when you scale them honestly.

Part IV — Machine learning and the SOTA quant frontier

27. The ML-in-trading taxonomy

Modern quantitative trading uses ML in several distinct ways:

1. Supervised return prediction. Given a feature vector (price history, accounting data, alternative data, etc.) for a security at time t, predict its return at time t+h. Tree ensembles (XGBoost, LightGBM, random forests) and neural networks dominate.
2. Supervised classification of regimes / events. Predict the probability of a regime change, earnings surprise, default, etc.
3. Unsupervised clustering / dimensionality reduction. PCA, autoencoders, k-means, t-SNE for regime identification, factor discovery, or risk-model construction.
4. Reinforcement learning. Optimal execution, market making, dynamic hedging — the action-reward structure of trading naturally fits the RL frame.
5. Generative models. GANs and diffusion models for synthetic market scenarios used in stress testing and data augmentation.
6. NLP / LLMs. Sentiment extraction from text (news, earnings transcripts, social media). Direct LLM-based forecasts (Lopez-Lira & Tang 2023).
7. Causal ML. Double machine learning (Chernozhukov et al. 2018) and related methods for clean estimation of treatment effects in observational financial data.

28. Marcos López de Prado's Advances in Financial Machine Learning (2018)

Lopez de Prado's book (Wiley 2018, "AFML") is the de-facto playbook for applying ML to financial data in a way that avoids the most common failure modes. Key contributions:

• Sample uniqueness. In finance, samples overlap: a 10-day forward return from yesterday and a 10-day forward return from today share 9 days. Standard cross-validation treats them as independent, dramatically over-fitting. Lopez de Prado proposes computing each observation's uniqueness (1 − overlap fraction) and weighting accordingly.
• Purged k-fold cross-validation. Standard k-fold CV leaks information from train to test when labels overlap with predictors. Purged CV removes training observations whose labels overlap with test labels. (Wikipedia: Purged cross-validation)
• Embargoed CV. Beyond purging, embargo a gap of h observations around each test fold to prevent serial-correlation leakage.
• Combinatorial purged CV (CPCV). Run many train/test splits combinatorially, producing thousands of out-of-sample evaluations rather than k. Used to compute the probability of backtest overfitting.
• The triple-barrier method. A labeling scheme for supervised learning on financial data: place upper, lower, and time barriers; label each sample by which barrier hits first.
• Meta-labeling. Use one primary model to generate signals (e.g., a trend filter saying "buy"), then a secondary ML model to decide whether to take each signal based on confidence. Decouples direction from sizing.

Study next: AFML (the book), then Lopez de Prado's Machine Learning for Asset Managers (CFA Institute Research Foundation, 2020), then Causal Factor Investing (2023).

29. The deflated Sharpe ratio and backtest overfitting

The single most dangerous activity in quantitative trading is testing many strategies and reporting the best. If you test 1000 random strategies, the best one will look great by pure chance, even with no real signal.

Bailey-Lopez de Prado (2014) Deflated Sharpe Ratio (DSR). Adjusts the observed Sharpe ratio for (a) the number of strategies tried, (b) skewness and kurtosis of returns, (c) the variance of the Sharpe estimate. The deflated value is interpretable as a probabilistic claim: "what's the probability that the true Sharpe exceeds some threshold, given how much I searched?" (Bailey-Lopez de Prado 2014; Wikipedia: Deflated Sharpe)

Probability of backtest overfitting (PBO). Bailey, Borwein, Lopez de Prado, Zhu (2014) propose a CSCV (combinatorially symmetric cross-validation) approach that simulates the probability that the best-performing strategy in-sample underperforms the median strategy out-of-sample.

The practical implication: any quantitative shop without aggressive multiple-testing correction is fooling itself.

30. Alternative data

"Alt data" is the umbrella term for non-traditional inputs:

• Satellite imagery: car counts at retail parking lots (predicts same-store sales), nighttime lights (predicts economic activity), agricultural yields, oil-storage tank levels.
• Web scraping: pricing data, job postings, app reviews, product availability.
• Credit-card panel data: spending at major retailers, sometimes daily.
• Geolocation data: foot traffic to stores, factory utilization.
• App-usage data: DAU/MAU for tech companies.
• Shipping: container traffic, AIS vessel tracking.
• Earnings call transcripts: NLP for tone, topic, executive turnover.
• Patent filings, FDA approvals, SEC EDGAR filings.

The peak alt-data hype was around 2017–2019. Empirically, alpha decays rapidly post-adoption — the moment 50 funds buy the same credit-card panel from Yipit or Earnest, the signal is largely arbitraged out within months. The persistent winners have been data providers themselves (Bloomberg, Refinitiv, S&P Market Intelligence, Yipit, M Science).

31. NLP and LLMs in trading

Three generations of text-driven signals:

1. Bag-of-words sentiment (RavenPack, Thomson Reuters MarketPsych, Bloomberg sentiment scores). Dictionary-based: count positive vs. negative words. Works mildly; well-arbitraged by mid-2010s.
2. Transformer-based domain-specific models: FinBERT (Yang et al. 2020) fine-tuned BERT on financial text. Better than BoW but still narrow.
3. General-purpose LLMs: GPT-3.5, GPT-4, Claude, Llama family. Lopez-Lira and Tang (2023) showed GPT-4 prompted with news headlines achieves ~90% hit-rate on directional same-day reaction and significant predictive ability for next-day drift, especially for small caps and bad news. They also document adoption-driven alpha decay over time. (Lopez-Lira & Tang 2023)

A 2024 follow-up paper, What Does ChatGPT Make of Historical Stock Returns? Extrapolation and Miscalibration in LLM Stock Return Forecasts (arXiv:2409.11540), shows that LLM-generated return forecasts exhibit systematic biases — extrapolation of recent returns and miscalibrated uncertainty — that traders need to correct for.

Practical use today. Most major hedge funds run LLMs over earnings transcripts, central-bank speeches, and 8-K filings for incremental signal. The signal is incremental, not transformative; LLMs are most useful as feature extractors for downstream models, not as stand-alone alpha engines.

32. Reinforcement learning for execution and market making

RL is conceptually well-suited to trading: the agent observes state (order book, position, time-to-deadline), takes an action (place or cancel an order), and receives a reward (P&L minus risk-adjusted costs).

• JPMorgan's LOXM (2017, internal): an RL-driven execution agent for equities. Reportedly outperforms traditional algos in cost/slippage on a substantial fraction of order flow.
• Deep Hedging (Buehler, Gonon, Teichmann, Wood 2019): trains a neural network to hedge derivative portfolios under realistic frictions (transaction costs, liquidity constraints, risk limits). The trained policy can handle exotic payoffs without analytic Greeks and outperforms Black-Scholes-style delta hedging when assumptions are violated. (Buehler et al. 2019)
• Market-making RL: numerous papers (Spooner-Savani 2018, Ganesh et al. 2019 at JPM, Sadighian 2019) show RL agents can match or exceed Avellaneda-Stoikov baselines, especially in markets with non-Brownian order arrivals.

The honest summary: RL is now a production technology at major banks and hedge funds for execution, hedging, and increasingly market making. It is not the silver bullet for alpha generation that the early hype suggested.

33. Generative models for synthetic market data

Backtests are limited by the historical data available. Generative adversarial networks (GANs) and more recently diffusion models can synthesize plausible additional market histories for stress testing and out-of-sample evaluation.

• QuantGAN (Wiese et al. 2020): a temporal convolutional GAN that reproduces stylized facts of equity returns (heavy tails, vol clustering, leverage effect).
• Sig-Wasserstein GANs (Ni et al. 2021): use signature-based metrics to match path properties.
• Diffusion models for time series: 2023–2024 work from JPMorgan AI Research and academic groups extends the image-generation diffusion machinery to multi-asset price paths.

The limitation: synthetic data inherits whatever distributional regularities the training set contained; it cannot generate regime changes the model has never seen.

34. Causal ML for finance

Standard ML in finance answers "given these features, what return is predicted?" — a predictive question. Causal ML answers "if I changed X, what would Y do?" — a counterfactual question. The distinction matters for understanding why a strategy works.

• Double / debiased machine learning (Chernozhukov, Chetverikov, Demirer, Duflo, Hansen, Newey, Robins 2018) provides ML-based estimators of treatment effects that are robust to high-dimensional nuisance parameters.
• Causal forests (Wager-Athey 2018) for heterogeneous treatment effects: e.g., does the impact of a Fed surprise depend on the credit cycle?

Applications in finance are nascent but growing — causal attribution of alpha, event studies, regulatory impact assessment.

35. Modern factor models: Bryan Kelly's research program

Yale's Bryan Kelly and coauthors have largely defined the machine-learning frontier in empirical asset pricing.

• Empirical Asset Pricing via Machine Learning (Gu, Kelly, Xiu 2020, RFS). Compared OLS, ridge, LASSO, elastic net, gradient-boosted trees, random forests, and neural networks on a panel of ~30k US stocks × ~60 years. Tree models and neural networks achieved roughly double the out-of-sample R² and equivalent Sharpe doubling vs. linear models. The most important predictors were variations on momentum, liquidity, and volatility. (Gu-Kelly-Xiu 2020)
• Autoencoder Asset Pricing Models (Gu, Kelly, Xiu 2021, Journal of Econometrics). Extends factor models to allow latent factors whose loadings are non-linear functions of firm characteristics. Outperforms Fama-French 5-factor and Kelly-Pruitt-Su (2019) "instrumented PCA" on out-of-sample pricing of US equities. (AQR working paper version)
• Characteristics are covariances (Kelly, Pruitt, Su 2019, JFE). Shows that the same stock characteristics that proxy for expected returns (in factor regressions) also proxy for factor exposures, providing a unified treatment.

The implication: the machine-learning factor model is not a refutation of Fama-French; it is the next step in the same research program, accommodating non-linearities and interactions that linear regressions miss.

36. Modern risk management

Value-at-Risk (VaR). The q-quantile loss over horizon h: "with probability q, your loss over horizon h will not exceed VaR." Almost always reported as 95% or 99% over 1-day or 10-day horizons.

Worked discrete VaR. Suppose your daily P&L distribution has the following discrete outcomes: -$100k with probability 0.01, -$50k with probability 0.04, -$20k with probability 0.10, +$5k with probability 0.55, +$20k with probability 0.30.

The 95% VaR is the smallest V such that Prob(Loss > V) ≤ 5%. Total prob of losing ≥ $50k = 0.01 + 0.04 = 0.05; total prob of losing ≥ $20k = 0.15. The 95% VaR = $50k.

The 99% VaR = $100k (probability of exceeding = 0.01).

Expected Shortfall (ES, aka CVaR or Tail VaR). The expected loss given that the loss exceeds the VaR threshold.

For the above distribution, 95% ES = E[Loss | Loss ≥ $50k] = (0.01 × $100k + 0.04 × $50k) / 0.05 = $3k/0.05 / 1 = ($1k + $2k) / 0.05 = $60k.

ES has three advantages over VaR:

1. Sub-additivity: the ES of a combined portfolio is ≤ sum of individual ESs. VaR can violate this (diversification can increase VaR for fat-tailed distributions).
2. Tail information: ES tells you how bad it gets when it does get bad; VaR is silent on what happens beyond the threshold.
3. Regulatory preference: Basel's FRTB (Fundamental Review of the Trading Book), in force in the EU and UK from January 2025, uses 97.5% ES rather than 99% VaR. (BPI on FRTB ES; AnalystPrep on FRTB)

Stress testing. Beyond statistical VaR/ES, regulators (CCAR in the US, EBA stress tests in Europe) require banks to simulate P&L under prescribed scenarios (e.g., "severe global recession with credit spreads doubling and equities falling 50%"). Internal stress tests at hedge funds run hundreds of historical and hypothetical scenarios.

Factor risk decomposition. Decompose portfolio variance into contributions from factors (market, sector, style). Tools: Barra (now MSCI Barra), Axioma, Bloomberg PORT, internal factor models.

Copulas. A flexible way to model joint distributions of multiple risks separately from each marginal distribution. Gaussian copulas (David Li's notorious 2000 paper) were blamed (rightly or not) for understating correlated default risk in CDO modeling pre-2008.

37. Backtesting metrics: a precise glossary

Sharpe ratio.

Sharpe = (mean excess return) / (standard deviation of excess return)

Typically annualized: multiply by √(periods per year). A "good" strategy has Sharpe > 1; a "great" one > 2; > 3 in production is rare and usually means hidden risk or overfit.

Sortino ratio. Like Sharpe but penalizes only downside volatility:

Sortino = (mean excess return) / (downside deviation)

where downside deviation is the standard deviation of returns below a threshold (often zero).

Calmar ratio.

Calmar = (annualized return) / |max drawdown|

Maximum drawdown (MDD). The largest peak-to-trough decline in cumulative wealth.

Worked example. Monthly cumulative returns: 100, 110, 120, 115, 100, 95, 105, 115, 130, 125. The peak so far at each point: 100, 110, 120, 120, 120, 120, 120, 120, 130, 130. Drawdown from each peak: 0, 0, 0, -4.2%, -16.7%, -20.8%, -12.5%, -4.2%, 0, -3.8%. Max drawdown = -20.8% (the trough at 95 from peak of 120).

Information ratio.

IR = (mean active return) / (tracking error)

Active return = portfolio return − benchmark return; tracking error = std of active returns.

Hit ratio / win rate. Fraction of profitable trades. A trend-follower may have hit rate < 50% but still profitable; a market-maker may have hit rate > 90% but blow up on adverse selection.

Profit factor.

PF = (sum of profits on winning trades) / (sum of losses on losing trades)

Kelly criterion. Given an edge with known win probability p and win/loss ratio b, the bet fraction that maximizes long-run logarithmic wealth is:

f* = (b·p − (1 − p)) / b

Worked example. A signal wins 55% of the time and pays 1:1 on each win (b = 1). Full-Kelly fraction is (1·0.55 − 0.45)/1 = 0.10, i.e., bet 10% of capital per trade. In practice, almost no one runs full Kelly — the optimal fraction is exquisitely sensitive to estimation error in p and b, and any over-estimate produces ruinous drawdowns. "Half-Kelly" or "quarter-Kelly" sizing is standard.

Regime detection. A live frontier of quantitative research is identifying which "regime" the market is in (low-vol vs. high-vol; risk-on vs. risk-off; trending vs. mean-reverting). Methods range from hidden Markov models (Hamilton 1989) and Markov-switching GARCH to neural-network classifiers on top of macro and microstructure features. The honest summary: regime detection is easy in-sample (you can always find labels that look meaningful), but extremely hard out-of-sample — the regimes that matter are the ones the model has never seen.

Part V — Crypto and DeFi market structure

38. CEX vs DEX: two market designs

Crypto trading happens on two distinct rails:

• Centralized exchanges (CEXs): Binance, OKX, Coinbase, Kraken, Bybit, Bitget. These look architecturally identical to traditional electronic exchanges: a limit order book, a matching engine, custodial settlement on the exchange's own balance sheet.
• Decentralized exchanges (DEXs): on-chain protocols that match trades algorithmically against pooled liquidity (AMMs) or against order books written to chain or to L2 systems (dYdX, Hyperliquid, Aevo).

CEX still dominates volume, but the DEX share has grown dramatically — Hyperliquid alone routinely trades $5–15B/day in perpetual swaps in 2025–2026, comparable to mid-tier CEX volume.

39. Constant-product AMMs (Uniswap V2)

Uniswap V2 launched in May 2020 and codified the "constant product market maker" (CPMM):

x · y = k

where x and y are the pool's reserves of the two assets, and k is constant (modulo fees). To buy Δx of token X, you must deposit enough Y to keep k constant:

(x − Δx) · (y + Δy) = x · y
⇒ Δy = (x · y) / (x − Δx) − y = (Δx · y) / (x − Δx)

Worked example. Pool has 100 ETH and 200,000 USDC. k = 20,000,000. You want to buy 10 ETH. New ETH balance = 90. Required USDC = 20,000,000 / 90 − 200,000 = 222,222 − 200,000 = 22,222 USDC. Effective price = 22,222/10 = 2,222 USDC/ETH, vs. the marginal pool price of 200,000/100 = 2,000 — a 11% price impact.

Spot price.

P = y / x

This is the marginal price at which an infinitesimal trade clears.

Impermanent loss (IL). If the external market price diverges from the pool's initial price, the LP holds less of the appreciated asset (because arbitrageurs have traded against the pool to bring it into line) than they would have if they'd just held both assets. The loss is "impermanent" because if the price returns, so does the position.

For a 50/50 V2 pool, with price ratio d = (new price) / (old price), the IL formula is:

IL = (2 · √d) / (1 + d) − 1

Worked example. Initial ETH = $1000. New ETH = $2000. d = 2. IL = (2·√2)/(1+2) − 1 = 2.828/3 − 1 = 0.943 − 1 = −5.7% vs. just holding 50/50. (Speedrun Ethereum: IL math)

Fees offset IL. LPs earn the swap fee (0.3% in V2) on every trade. In sufficiently active pools, fees can dominate IL — but in any meaningful price divergence, IL wins.

40. Concentrated liquidity (Uniswap V3, May 2021)

V3 lets LPs concentrate liquidity in a price range [P_lower, P_upper]. Within the range, the position behaves like a V2 position with a virtual reserve much larger than actually deposited (capital efficiency 10x–4000x depending on range). Outside the range, it becomes 100% one asset and earns no fees. (Uniswap V3 explained; Uniswap support: IL)

Trade-off: higher fees within the range, but IL is amplified (effectively leveraged) and you earn nothing outside the range. Active LPs rebalance ranges continually; passive LPs typically lose money to skilled active LPs and to JIT (just-in-time) liquidity providers who place narrow ranges only when large trades are pending in the mempool.

41. Other AMM curves

• Curve Finance's StableSwap uses a hybrid curve that's almost flat near the equal-ratio point (low slippage for similar-priced assets like USDC/USDT/DAI) and reverts to constant-product at the edges. Equation involves a bonding curve with a parameter A controlling the steepness.
• Balancer: generalized weighted-product CPMM with arbitrary weights and ≥2 assets.
• Trader Joe Liquidity Book: discretized concentrated liquidity in "bins," each with constant-sum within the bin.
• Curve V2 / Cryptoswap: combines stableswap with concentrated-liquidity-like rebalancing for volatile pairs.

The mathematical theme: AMMs are convex bonding curves. The choice of curve trades capital efficiency, slippage, and IL.

42. Perpetual swaps (perps)

Perpetual futures are the dominant derivative instrument in crypto. They are futures without expiry, kept tethered to spot via a periodic funding rate payment between longs and shorts.

Funding rate mechanism. If the perp trades above spot (the mark-index basis is positive), longs pay shorts; if below, shorts pay longs. The mechanic creates the economic incentive that drags the perp price back toward spot.

Funding rate formula (simplified, Binance-style).

Funding Rate = Premium Index + clamp(Interest Rate − Premium Index, -0.05%, 0.05%)

where the Premium Index measures the perp-spot deviation over the funding interval and the Interest Rate is a base rate (often 0.01%/8h).

Funding is typically paid every 8 hours (Binance, OKX, dYdX), every hour (some venues), or continuously (Hyperliquid: every second at a small per-second rate that cumulates to roughly the 1h or 8h equivalents). (Hyperliquid via 21Shares; DeFi-explained: funding rates)

Worked funding example. ETH perp trades at $3,015. ETH spot at $3,000. Premium = (3015 − 3000)/3000 = 0.5%. Interest rate = 0.01%. Clamped term = clamp(0.01% − 0.5%, -0.05%, 0.05%) = -0.05%. Funding rate = 0.5% − 0.05% = 0.45% for this 8h interval.

A long with $100,000 notional pays $450 to the short side; a short receives $450. Annualized funding = 0.45% × 3 funding intervals/day × 365 = ~493%.

(This is why crypto perp carry can be enormous in either direction during euphoria or panic, and why "delta-neutral basis trades" — long spot, short perp, collect funding — are a popular structured product.)

Three major perp DEXs.

• dYdX V4 (Cosmos-based app-chain, launched late 2023): a true on-chain order book with off-chain matching, settling on its own L1. Funding settled every 8 hours. (Bitcoin.com: perp DEXs 2026)
• GMX (Arbitrum, launched 2021): no order book; uses a multi-asset GLP/GM liquidity pool that LPs deposit into and that traders take the other side of. Uses a borrow-fee model rather than a classical funding rate (the heavier side of open interest pays the lighter side an hourly fee scaled by pool utilization).
• Hyperliquid (custom L1, launched 2023, has dominated 2024–2026 perp volume): fully on-chain order book with sub-second blocks. Continuous funding (per-second accrual). Capable of millions of TPS for trading-specific workloads. Token (HYPE) launched November 2024 with a major airdrop; founder-led, no VCs. By 2025 routinely captured >70% of decentralized perp volume.

43. MEV: maximal extractable value

MEV is the profit extractable by reordering, inserting, or censoring transactions in a block. The taxonomy:

• DEX arbitrage: when two AMMs offer different prices for the same pair, a searcher executes a flash-trade that arbitrages both. Profit ≈ the spread minus gas.
• Liquidations: in lending protocols (Aave, Compound, MakerDAO), under-collateralized positions can be liquidated for a bonus by anyone who calls the liquidation function. MEV searchers compete for these.
• Sandwich attacks: a searcher detects a pending user swap in the mempool, places a buy before it (pushing the price up), then sells after the user's swap (collecting the user's slippage as profit). The user pays a worse price; the searcher earns the difference.
• Just-in-time (JIT) liquidity: in V3, a sophisticated LP places a narrow range exactly bracketing a pending swap, collects the fee, and withdraws. Sometimes welfare-enhancing for the trader (better execution), sometimes parasitic on passive LPs.

Flashbots and MEV-Boost. Flashbots is a research org founded in 2020 to mitigate MEV's harms. Their key innovation: a private mempool (Flashbots Auction) where searchers submit transaction bundles directly to miners/validators, paying for inclusion. This prevents front-runners from observing pending searcher activity and prevents adversarial reorderings. (Flashbots overview)

After Ethereum's transition to Proof of Stake, MEV-Boost became the de facto middleware: validators outsource block construction to "builders" who compete to assemble the most profitable block. The validator simply selects the highest-paying block. As of 2024–2026, well over 80% of Ethereum blocks are built via MEV-Boost. (WeufoundationL MEV/Flashbots/SUAVE)

Proposer-Builder Separation (PBS). The formalization of the MEV-Boost architecture into Ethereum's protocol roadmap. Goal: keep validators (the proposers) from needing direct MEV expertise, concentrating sophistication in a competitive builder market. The downside: builder centralization (a small number of builders win most blocks; the top 3 build ~80%). (BlockBase: MEV & PBS; SwapSpace overview)

The size of MEV. Ethereum MEV in 2024 was roughly $400–700M extracted, depending on how you count. Solana MEV (typically much higher because of lower-fee high-volume DEX activity) is harder to measure but likely larger in dollar terms.

44. Crypto market structure peculiarities

• 24/7 trading: no closing bell, no weekend halts. Liquidity is thinnest Sunday US-evening through Monday morning Asia.
• Stablecoin dynamics: USDT ($130B market cap), USDC ($60B), DAI/sDAI ($5B+), USDe ($7B+) are the de-facto trading currency. The "USDT depeg risk" is the perennial systemic question — short-lived depegs occurred in May 2022 (Luna collapse) and March 2023 (USDC briefly broke during the SVB crisis).
• Fragmented venues: even within a single asset, prices can differ noticeably across exchanges. CEX-to-CEX arbitrage was a Renaissance-style cash cow 2017–2020 and is still profitable, though much tighter.
• Basis trades: long spot, short futures (or short perp) at a premium. The "cash-and-carry" of crypto. During bull cycles, basis annualized 15–40% has been common; during bears, it can flip negative.
• Perp-spot arbitrage: hold spot Bitcoin, short the same notional in perp futures, collect funding. Run at scale by Ethena (sUSDe) and similar protocols, with substantial AUM.
• On-chain analytics: Glassnode, Coin Metrics, Nansen, Chainalysis, Arkham — the "alternative data" of crypto. Inputs include exchange in/outflow, miner positioning, long-term holder behavior, smart-money wallet tracking. Some signals appear to have predictive power (e.g., exchange outflows are bullish), but the alt-data alpha-decay problem applies here too.

Part VI — Live debates and frontier issues (2024–2026)

45. The 0DTE options explosion

Zero-days-to-expiration options (options that expire the same day they are traded) have become the dominant slice of US index options.

• Volume: 0DTE represented ~5% of SPX option volume in 2016, ~17% in 2020, ~43% in 2023, and roughly 50% in 2024–early 2025. (Cboe Volatility Insights; Optionalpha trends)
• Why now: Cboe added daily SPX expirations on Tuesdays and Thursdays in May 2022, completing the Monday-through-Friday daily schedule. Pre-2022, weekly expirations existed but daily did not.
• Who trades 0DTE: split roughly between sophisticated institutional flow (dispersion, gamma-hedging, vol harvesting) and rapidly-growing retail (Robinhood, tastytrade, Webull, Schwab).
• Market impact debate: Critics argue 0DTE flow amplifies intraday moves through dealer hedging. Cboe research argues 0DTE flow is more balanced than critics claim and not systematically destabilizing. The truth depends on dealer net gamma positioning, which varies day-to-day. (Numerix; Resonanz: 0DTE institutional lessons)

Practical implication: 0DTE has shifted the SPX market's effective microstructure. A trader assessing index conditions in 2025 needs to know dealer gamma exposure as a baseline, the way bond traders in 2010 needed to know primary-dealer SOMA holdings.

46. Passive investing's market-structure effects

Roughly 50–55% of US equity fund assets are now in index/ETF passive products. The big questions:

• Does passive investing degrade price efficiency? Coles, Heath and Ringgenberg (2022) find no significant impact on price informativeness from exogenous shocks to indexing — active investors fill the gap as needed. (Coles-Heath-Ringgenberg on SSRN; ScienceDirect) Jiang, Vayanos and Zheng (2023, "Passive Investing and the Rise of Mega-Firms") argue the opposite: passive flows mechanically inflate the largest stocks more than the rest, contributing to mega-cap concentration. (LSE PDF)
• Are flows reflexive? Petajisto (2011) showed about half of S&P 500 inclusion price impact reverses over the following two months — flow can move prices temporarily even when no fundamental information has changed.
• Who owns the corporate vote? The "Big Three" (BlackRock, Vanguard, State Street) collectively vote ~25% of S&P 500 shares. Their proxy-voting behavior is a live corporate-governance issue.

The honest summary: passive investing has reshaped flows and concentration, but its impact on price efficiency narrowly defined appears modest, because active arbitrageurs scale to maintain pricing in equilibrium.

47. The private credit boom

Private credit AUM has grown from $158B in 2010 to nearly $2T globally by mid-2024. US private credit assets reached ~$1.34T by mid-2024. The growth has been driven by:

• Bank-regulation tightening (Basel III, post-Dodd-Frank leveraged-lending guidance) pushed risky middle-market lending out of banks.
• Private equity sponsorship of private-credit-backed buyouts in lieu of broadly syndicated loans.
• Investor demand for higher yields than public credit offered through the low-rate era.

Implications for trading: a growing pool of opaque, illiquid credit assets sits outside the daily mark-to-market discipline of public markets. Stress would show up first in PE valuations, then in fund-level redemption queues, then potentially in BDC equity (the publicly listed Business Development Companies that are the most visible window into private credit). The Federal Reserve, FDIC, and IMF have all flagged the sector. (Federal Reserve Bank of Boston; Fed Notes May 2025)

48. Retail trading post-Robinhood

The 2020–2021 retail surge (zero-commission via Robinhood, stimulus checks, lockdown boredom, GameStop) reshaped equity market participation. Persistent effects:

• Retail share of US equity volume remains structurally higher (~20–25%) than pre-2020 (~10–15%).
• Single-stock and 0DTE options have become a major retail asset class.
• "Memecoins" and meme-stocks demonstrated the power of coordinated social-media-driven flow.
• The GameStop episode showed that even sophisticated short-sellers (Melvin Capital) can be destroyed by a sufficiently coordinated retail squeeze in a highly-shorted name.

49. AI in trading firms — hype vs reality

A sober scorecard:

• Production use today: NLP feature extraction from text, RL-driven execution (LOXM and equivalents), deep hedging (production at JPM, Goldman, BNP), ML-based risk model tuning, alt-data signal extraction, regime classification, exception detection in surveillance.
• Plausibly real but unverified: end-to-end LLM-driven systematic trading at small experimental scale. Stories of "agentic" trading AIs running entire portfolios are largely marketing.
• Mostly hype: claims that LLMs alone produce alpha at institutional scale. The Lopez-Lira papers show real but small and decaying effects. Anyone selling you a "GPT trading bot" with high Sharpe is selling backtest overfitting.

50. Crypto-equity correlation evolution

The 2020–2022 era saw Bitcoin trade as a high-beta tech-stock proxy: 30-day rolling correlation with the S&P 500 frequently exceeded 70%. (Phemex correlation analysis)

Through 2024–2025, the correlation became unstable: by Q3 2025 it had dropped near zero, then spiked again to ~70% during early-2025 risk-off episodes. The five-year average sits around 0.38. Bitcoin's daily volatility runs 3–5× the S&P 500's, so even moderate correlations imply substantial beta. (Ainvest: BTC correlation analysis; Stoic.ai BTC vs S&P 500)

The Bitcoin spot ETF approvals (January 2024) accelerated financialization — Bitcoin is now a portfolio asset for many institutional allocators, which mechanically links its flows to broader risk sentiment.

51. Current state of factor performance

• Value: a decade-long drought from ~2010–2020 followed by a strong recovery starting November 2020 with the Pfizer vaccine. Value has performed reasonably in 2021–2024; 2025 has been mixed. (Robeco: value's "death greatly exaggerated")
• Momentum: continues to work well in time-series form (trend-following had its best decade since the 1990s in 2022–2024). Cross-sectional momentum more cyclical.
• Quality / profitability: consistent positive returns through the post-2020 period. The most robust factor of the modern era.
• Low volatility: positive but modest; sensitive to interest-rate regime.
• Size: weakest of the canonical factors; many practitioners argue it doesn't survive after micro-cap and trading-cost corrections.

Part VII — Practitioner archetypes: brief profiles

These short profiles are reference material; they are deliberately concise.

Renaissance Technologies (East Setauket, NY). Founded by Jim Simons in 1982. The Medallion Fund (closed to outside money since 1993; >$10B in size, capped to preserve alpha) has generated ~66% gross / ~39% net annualized returns since 1988. Quant-only. Recruits from math, physics, computer science, speech recognition; minimal Wall Street résumés. Strategy: thousands of short-horizon statistical signals across futures, equities, and FX. P-value thresholds reportedly < 0.01. (Quartr: Renaissance breakdown; [Zuckerman, The Man Who Solved the Market, 2019])

Two Sigma (NYC). Founded by John Overdeck and David Siegel (both ex-D.E. Shaw) in 2001. ~$60B AUM. Quant-and-discretionary hybrid; substantial venture investing arm. Known for engineering culture and academic publications (the Two Sigma research site is a public-facing showcase).

D.E. Shaw (NYC). Founded by David Shaw (Columbia CS faculty) in 1988. ~$60B AUM. The first "quant hedge fund" in the modern sense; alumni include Jeff Bezos, Lawrence Stein, John Overdeck, David Siegel. Mixed quant and discretionary strategies; computational biology and energy desks.

Citadel and Citadel Securities (Chicago/Miami). Citadel hedge fund (Ken Griffin, founded 1990, ~$70B AUM) is a multi-strategy operation across equities, fixed income, commodities, credit, and quant. Citadel Securities (separately a top-tier electronic market maker) handles ~25% of US equity volume. (Bloomberg: Citadel & Jane Street record revenue)

Jane Street (NYC, London, Hong Kong, Amsterdam). Founded by Tim Reynolds, Marc Gerstein, Rob Granieri, and Michael Jenkins in 2000. Partnership; tightly-held, ~3,000 employees. Dominant in ETF market-making, expanding aggressively into bonds and crypto. $20.5B net trading revenue in 2024; took >10% of North American equity volume. Known for OCaml infrastructure and an idiosyncratic culture. (Global Trading: Jane Street 10% of US equity; Disruption Banking on Jane Street)

Jump Trading (Chicago). Founded 1999. Ultra-low-latency electronic market maker. Major crypto presence via Jump Crypto. Famously private; very few public talks or papers.

Hudson River Trading (NYC). Latency arbitrage and electronic market-making across equities, futures, options, FX, ETFs.

XTX Markets (London). Founded 2015 by Alex Gerko (ex-Deutsche Bank). Pure-quant electronic market maker; FX-first, expanded into equities and futures. Privately held; ~$10B+ annual revenue.

Optiver, IMC, Susquehanna (SIG), DRW, Tower Research. The other major electronic market-making firms. SIG (Susquehanna) is unusual for its game-theory training program — new traders famously learn poker before they learn options.

AQR (Greenwich). Cliff Asness, David Kabiller, Robert Krail, John Liew (1998). Academic-research-driven systematic: value, momentum, carry, quality, low vol. Suffered the 2018–2020 quant winter, recovered strongly 2022–2024. Asness, Frazzini, Pedersen, and Moskowitz are all publishing academics with deep ties to AQR.

Bridgewater (Westport, CT). Ray Dalio (1975, transitioned leadership to Bob Prince and Karen Karniol-Tambour). ~$160B AUM. Pure Alpha (discretionary macro), All Weather (risk parity), Optimal Portfolio. The most-followed daily research note in macro: Daily Observations.

Millennium (NYC). Izzy Englander (1989). ~$70B AUM. Multi-strategy "pod shop" model: hundreds of small portfolio-manager teams trade independently, with extremely tight risk limits.

Point72 (Stamford, CT). Steve Cohen, successor to SAC Capital (closed 2014 in an insider-trading settlement). ~$35B AUM. Discretionary equity long-short with growing systematic and quant components.

Tudor Investment Corp. Paul Tudor Jones (1980). Discretionary global macro; the 1987 crash short is his founding legend. ~$11B.

Brevan Howard (London). Alan Howard (2002). Discretionary macro. Had a major 2024 stumble (assets fell from ~$35B peak); 2025 saw stabilization.

Soros Fund Management. George Soros (1969). Famously broke the Bank of England in 1992 (~$1B+ shorting sterling out of the ERM). Now run by Dawn Fitzpatrick as a family office (Soros stopped accepting outside money in 2011).

Druckenmiller (Duquesne Family Office). Stanley Druckenmiller — former Soros, now manages his own capital with a small team. Public commentary remains influential.

Buffett (Berkshire Hathaway). Discussed above. ~$170B equity portfolio; controlling stakes in dozens of operating businesses.

Howard Marks (Oaktree, Los Angeles). Distressed debt and credit. Public Memos are required reading for credit investors.

Seth Klarman (Baupost, Boston). Value investor in the Graham-Dodd lineage. Wrote Margin of Safety (1991); rare and expensive secondhand.

Michael Burry (Scion Asset Management). The Big Short pre-2008. Continues to make highly contrarian, concentrated bets; public Twitter/X presence.

Bill Ackman (Pershing Square). Activist long-only. Notable trades: Herbalife short (lost), $26M-to-$2.6B credit-default-swap pandemic trade (March 2020), MBIA, Valeant (lost badly).

Mathematical foundations

This section consolidates the algebra used throughout. Use it as a reference.

Returns

• Simple return: R = (P_t − P_{t-1}) / P_{t-1}
• Log return: r = ln(P_t / P_{t-1}) = ln(1 + R)
• Log returns are additive across time; simple returns are additive across portfolios. Most academic work uses log; most practitioner P&L is simple.

Sharpe ratio

Sharpe = (mean(R) − r_f) / std(R)

Annualization: multiply by √(periods per year). For daily returns, √252 ≈ 15.87.

Portfolio mean and variance (two assets)

E[R_p] = w₁·E[R₁] + w₂·E[R₂]
σ²_p   = w₁²σ₁² + w₂²σ₂² + 2·w₁·w₂·ρ₁₂·σ₁·σ₂

N-asset version

E[R_p] = Σ_i w_i · E[R_i]
σ²_p   = Σ_i Σ_j w_i · w_j · σ_ij

where σ_ij is the covariance between assets i and j (σ_ii = σ_i²).

CAPM

E[R_i] = R_f + β_i · (E[R_m] − R_f)

β_i = Cov(R_i, R_m) / Var(R_m)

Worked beta. Daily returns for a stock and the market over 5 days:

• Stock: +1.0%, -0.5%, +2.0%, -1.5%, +0.5% (mean = +0.3%)
• Market: +0.5%, -0.3%, +1.2%, -1.0%, +0.4% (mean = +0.16%)

Deviations from mean:

• Stock: +0.7, -0.8, +1.7, -1.8, +0.2
• Market: +0.34, -0.46, +1.04, -1.16, +0.24

Cov ≈ (0.7·0.34 + 0.8·0.46 + 1.7·1.04 + 1.8·1.16 + 0.2·0.24)/5 ≈ (0.238 + 0.368 + 1.768 + 2.088 + 0.048)/5 ≈ 4.51/5 = 0.902

Var(market) ≈ (0.34² + 0.46² + 1.04² + 1.16² + 0.24²)/5 ≈ (0.116 + 0.212 + 1.082 + 1.346 + 0.058)/5 ≈ 2.81/5 = 0.563

β ≈ 0.902 / 0.563 ≈ 1.60

Factor regression

R_i − R_f = α + β_MKT·(R_m − R_f) + β_SMB·SMB + β_HML·HML + β_UMD·UMD + ε

α (the intercept) is the "abnormal return" after controlling for factor exposures. The factor loadings (β) tell you what factor tilts the portfolio has. Run an OLS or, for time-varying loadings, a Kalman filter / rolling regression.

Black-Scholes (recap)

C = S · N(d₁) − K · e^(−rT) · N(d₂)
P = K · e^(−rT) − S + C  (put-call parity)

d₁ = [ ln(S/K) + (r + σ²/2)·T ] / (σ·√T)
d₂ = d₁ − σ·√T

Binomial tree (one step)

p = (e^(rΔt) − d) / (u − d)         # risk-neutral up-probability
C = e^(−rΔt) · [p · C_u + (1−p) · C_d]

VaR and ES

For a discrete distribution of P&L outcomes:

• VaR at level α: smallest V such that P(Loss > V) ≤ 1 − α.
• ES at level α: E[Loss | Loss ≥ VaR_α].

For a continuous distribution with PDF f(x):

VaR_α = − F⁻¹(1 − α)
ES_α  = − (1 / (1 − α)) · ∫_{-∞}^{VaR_α} x · f(x) dx

Maximum drawdown

DD_t  = (max_{s ≤ t} W_s − W_t) / max_{s ≤ t} W_s
MDD   = max_t DD_t

where W_t is cumulative wealth at time t.

Sortino and Calmar

Sortino = (mean excess return) / √(E[(R − threshold)² | R < threshold])
Calmar  = (annualized return) / |MDD|

Information ratio

IR = mean(R_p − R_b) / std(R_p − R_b)

Constant-product AMM math

Pool invariant: x · y = k

Buy Δx of token X: pay Δy = (Δx · y) / (x − Δx)

Spot price: P = y / x

Impermanent loss (50/50 pool, price ratio d = P_new / P_old):
IL = 2·√d / (1 + d) − 1

Funding rate (perp)

Funding = Premium Index + clamp(Interest Rate − Premium Index, −cap, +cap)

Premium Index ≈ (Mark Perp Price − Index Spot) / Index Spot, averaged across funding interval.

Annualization: multiply per-interval rate by number of intervals per year.

Avellaneda-Stoikov optimal quotes (intuition only)

Maker chooses bid and ask to maximize:

E[U(X_T)] = E[X_T] − (γ/2) · Var[X_T]

where X is wealth, γ is risk aversion. The optimal solution gives:

• Reservation price: r = S − q · γ · σ² · (T − t) (shift away from mid in direction that unloads inventory q)
• Half-spread: a function of (γ, σ², T−t, order-arrival intensity κ).

Optimal bid = r − half-spread; optimal ask = r + half-spread.

Study next (for derivation): Hamilton-Jacobi-Bellman equations, dynamic programming in continuous time, viscosity solutions. The textbook is Cartea-Jaimungal-Penalva, Algorithmic and High-Frequency Trading.

Risks, limitations, and open questions

Risks of this report's framing.

• I have foregrounded the mainstream academic and practitioner consensus. Heterodox positions (post-Keynesian finance, MMT-inflected market analysis, radical critiques of EMH from a behavioral standpoint) are represented but not given equal weight.
• Survivorship bias is everywhere in narrative finance. The famous practitioners are famous because they succeeded; the comparable failures rarely write memoirs. Treat all "what made X great" stories as Bayesian updates with high variance.
• Most quoted backtested results in popular finance literature suffer from at least one of: data snooping, survivorship bias, ignoring trading costs, ignoring borrow costs, ignoring liquidity constraints, or overfitting to a single sample.

Open empirical questions.

• Does the value premium genuinely persist out-of-sample post-2020, or is the 2020-2024 rebound a one-time correction?
• Is the equity risk premium currently elevated, normal, or compressed? (Damodaran's implied-ERP page is the cleanest live tracker.)
• Does the apparent post-publication decay of academic factors mean factors vanish, or just shrink to the level of a competitive equilibrium return on capital?
• What is the true tail risk of crowded short-vol and basis trades? (Each blow-up so far — 2018 XIV, 2020 Treasury basis, 2022 LDI, 2022 TerraLuna — has been treated as idiosyncratic, but the family resemblance is hard to miss.)
• Can ML deliver genuine new signal or only smarter combinations of known signals? The Gu-Kelly-Xiu evidence suggests mostly the latter.
• What happens to crypto market structure if/when one or more of (USDT, perp-DEX dominance, Ethereum L2 fragmentation, MEV-Boost centralization) breaks?

Open structural questions.

• 0DTE dealer-gamma feedback: is there a regime where it triggers a flash crash equivalent to August 24, 2015 or May 6, 2010?
• Private credit: is the FDIC's 2025 concern justified, and does the lack of mark-to-market transparency hide a building loss spiral?
• Passive investing: does the Big Three's combined voting weight create governance distortions even if it doesn't (per Coles-Heath-Ringgenberg) distort prices?
• Hyperliquid: does on-chain order-book trading represent the future microstructure of crypto, or a high-volume novelty that fragments back to CEXs in the next bear market?

Things this report deliberately under-covers.

• Cross-border regulation (MiFID II/III in Europe, MAS/HKMA in Asia).
• Currency-specific peculiarities (CNH offshore market, EM frontier markets).
• Insurance-linked securities (catastrophe bonds, ILW).
• Carbon markets (EU ETS, voluntary markets).
• The full taxonomy of structured products (autocallables, accumulators).
• Niche cryptocurrencies and the L2 wars beyond the major players named.
• Detailed quantitative coverage of credit derivatives (CDS, CDX, CLO mechanics).

If you want depth in any of these, they merit dedicated companion reports.

Recommendations for further study

A staged textbook progression for the serious self-learner, ordered roughly by required mathematical maturity:

Stage 1: foundations (months 1–3)

1. Burton Malkiel — A Random Walk Down Wall Street. The popular distillation of EMH. Read first to set the baseline.
2. John Bogle — Common Sense on Mutual Funds. The case for passive investing, by the man who built Vanguard.
3. Benjamin Graham — The Intelligent Investor. Read with Jason Zweig's annotated edition. The single best non-quantitative book on investing temperament.
4. Howard Marks — The Most Important Thing. Distillation of his Oaktree memos. Best primer on risk thinking.
5. Ray Dalio — Principles for Navigating Big Debt Crises. Free PDF on Bridgewater's website. The macro thinking that built All Weather.

Stage 2: textbook fluency (months 3–9)

1. Bodie, Kane, Marcus — Investments. The standard MBA textbook. Covers portfolio theory, CAPM, factor models, fixed income, options, all at algebra level.
2. John Hull — Options, Futures and Other Derivatives. The canonical reference for derivatives. Chapters 1–14 are essential; the rest are reference.
3. Bruce Tuckman & Angel Serrat — Fixed Income Securities. The fixed-income equivalent of Hull. Carefully developed yield-curve and rates-derivative material.
4. Andrew Ang — Asset Management: A Systematic Approach to Factor Investing. Modern factor-investing textbook with empirical grounding.
5. Antti Ilmanen — Expected Returns. The most thorough survey of empirical expected returns across asset classes. Updated 2022 edition: Investing Amid Low Expected Returns.

Stage 3: quantitative toolkit (months 9–18)

1. Grinold & Kahn — Active Portfolio Management. The original "fundamental law of active management" framework: IR = IC · √breadth.
2. Marcos Lopez de Prado — Advances in Financial Machine Learning. The de-facto playbook for applying ML to financial data without fooling yourself.
3. Marcos Lopez de Prado — Machine Learning for Asset Managers. The shorter, more practitioner-oriented companion.
4. Ernie Chan — Algorithmic Trading: Winning Strategies and Their Rationale. Hands-on backtesting; pairs trading; mean reversion; momentum.
5. Robert Carver — Systematic Trading and Leveraged Trading. Practical systematic-trading guidance from an ex-AHL practitioner.

Stage 4: rigorous mathematics (months 18+, or in parallel)

1. Steven Shreve — Stochastic Calculus for Finance, Vols I & II. The single best route into Brownian motion, Itô calculus, the Black-Scholes derivation, and risk-neutral pricing. Vol I is binomial; Vol II is continuous.
2. Damir Filipović — Term-Structure Models. The mathematical theory of interest-rate modeling.
3. Cartea, Jaimungal, Penalva — Algorithmic and High-Frequency Trading. The graduate-level treatment of optimal execution, market making (including the full Avellaneda-Stoikov derivation), and HFT mathematics.
4. Olivier Guéant — The Financial Mathematics of Market Liquidity. Optimal execution and liquidity modeling at the research frontier.
5. Tomas Björk — Arbitrage Theory in Continuous Time. The pricing-by-no-arbitrage textbook.

Stage 5: current research

• Marcos Lopez de Prado, latest SSRN papers (~10/year on ML, backtesting, causal factor investing).
• Bryan Kelly's SSRN page for autoencoder asset pricing, machine-learning factors, transformer-based prediction.
• AQR Capital research notes (free): Asness, Frazzini, Pedersen, Israel, Ilmanen. Regularly publishing.
• Two Sigma Insights (free, online): high-quality applied-research notes.
• BIS Quarterly Review for market structure, central bank, and cross-border-banking research.
• Federal Reserve Liberty Street Economics blog for primary-source Fed thinking on US markets.

Stage 6: crypto-specific reading

• Paradigm Research blog for technical AMM, MEV, and L2 research.
• a16z Crypto Research for protocol design and tokenomics.
• Vitalik Buterin's blog for Ethereum and PBS roadmap thinking.
• Flashbots research (writings.flashbots.net) for MEV.
• Galaxy Research and Coin Metrics State of the Network for institutional-quality market structure.
• The original Uniswap V2 and V3 whitepapers, the dYdX V4 docs, and Hyperliquid's published documentation.

Stage 7: broadening (always)

• Matt Levine — Money Stuff daily newsletter (Bloomberg, free with sign-up). The best ongoing market commentary in English.
• FT Alphaville for daily institutional finance.
• Patrick O'Shaughnessy — Invest Like the Best podcast for long-form practitioner interviews.
• Odd Lots podcast (Joe Weisenthal & Tracy Alloway) for market-structure deep dives.
• Newfound Research / Corey Hoffstein for thoughtful systematic investing commentary.
• Climateer Investing, A Wealth of Common Sense (Carlson), Bronte Capital (John Hempton) for thoughtful longform.
• Robin Wigglesworth — Trillions for the index-investing history; his FT writing for ongoing coverage.

Source inventory

Below is a categorized list of the principal sources used. Citation discipline: every major claim in the body either links directly to a source or is supported by an item in this inventory.

Foundational academic papers

• Markowitz, H. (1952). "Portfolio Selection." Journal of Finance 7(1), 77–91.
• Sharpe, W. (1964). "Capital Asset Prices: A Theory of Market Equilibrium Under Conditions of Risk." Journal of Finance 19(3), 425–442.
• Black, F., & Scholes, M. (1973). "The Pricing of Options and Corporate Liabilities." Journal of Political Economy 81(3), 637–654.
• Merton, R. (1973). "Theory of Rational Option Pricing." Bell Journal of Economics and Management Science 4(1), 141–183.
• Fama, E. (1970). "Efficient Capital Markets: A Review of Theory and Empirical Work." Journal of Finance 25(2), 383–417. https://www.bu.edu/econ/files/2011/01/Fama2.pdf
• Grossman, S. & Stiglitz, J. (1980). "On the Impossibility of Informationally Efficient Markets." American Economic Review 70(3), 393–408. https://www.aeaweb.org/aer/top20/70.3.393-408.pdf
• Kyle, A. (1985). "Continuous Auctions and Insider Trading." Econometrica 53(6), 1315–1335. https://personal.utdallas.edu/~nina.baranchuk/Fin7310/papers/Kyle1985.pdf
• Glosten, L. & Milgrom, P. (1985). "Bid, ask and transaction prices in a specialist market with heterogeneously informed traders." Journal of Financial Economics 14(1), 71–100.
• Fama, E. & French, K. (1993). "Common Risk Factors in the Returns on Stocks and Bonds." Journal of Financial Economics 33(1), 3–56.
• Carhart, M. (1997). "On Persistence in Mutual Fund Performance." Journal of Finance 52(1), 57–82. https://onlinelibrary.wiley.com/doi/10.1111/j.1540-6261.1997.tb03808.x
• Almgren, R. & Chriss, N. (2000). "Optimal Execution of Portfolio Transactions." Journal of Risk 3, 5–39. https://www.smallake.kr/wp-content/uploads/2016/03/optliq.pdf
• Lo, A. (2004). "The Adaptive Markets Hypothesis: Market Efficiency from an Evolutionary Perspective." Journal of Portfolio Management 30(5), 15–29. https://web.mit.edu/Alo/www/Papers/JPM2004_Pub.pdf
• Avellaneda, M. & Stoikov, S. (2008). "High-frequency Trading in a Limit Order Book." Quantitative Finance 8(3), 217–224.
• Moskowitz, T., Ooi, Y. & Pedersen, L. (2012). "Time Series Momentum." Journal of Financial Economics 104(2), 228–250. http://docs.lhpedersen.com/TimeSeriesMomentum.pdf
• Bailey, D. & Lopez de Prado, M. (2014). "The Deflated Sharpe Ratio." Journal of Portfolio Management. https://www.davidhbailey.com/dhbpapers/deflated-sharpe.pdf
• Frazzini, A. & Pedersen, L. (2014). "Betting Against Beta." Journal of Financial Economics 111(1), 1–25.
• Fama, E. & French, K. (2015). "A Five-Factor Asset Pricing Model." Journal of Financial Economics 116(1), 1–22. https://tevgeniou.github.io/EquityRiskFactors/bibliography/FiveFactor.pdf
• Hou, K., Xue, C. & Zhang, L. (2015). "Digesting Anomalies: An Investment Approach." Review of Financial Studies 28(3), 650–705. https://academic.oup.com/rfs/article/28/3/650/1574802
• Harvey, C., Liu, Y. & Zhu, H. (2016). "...and the Cross-Section of Expected Returns." Review of Financial Studies 29(1), 5–68. https://academic.oup.com/rfs/article/29/1/5/1843824
• Stambaugh, R. & Yuan, Y. (2017). "Mispricing Factors." Review of Financial Studies 30(4), 1270–1315. https://academic.oup.com/rfs/article/30/4/1270/2965095
• Koijen, R., Moskowitz, T., Pedersen, L. & Vrugt, E. (2018). "Carry." Journal of Financial Economics 127(2), 197–225. https://research.cbs.dk/files/57294842/lasse_heje_pedersen_et_al_carry_acceptedmanuscript.pdf
• Asness, C., Frazzini, A. & Pedersen, L. (2019). "Quality Minus Junk." Review of Accounting Studies 24(1), 34–112.
• Buehler, H., Gonon, L., Teichmann, J., Wood, B., et al. (2019). "Deep Hedging." Quantitative Finance 19(8), 1271–1291.
• Gu, S., Kelly, B. & Xiu, D. (2020). "Empirical Asset Pricing via Machine Learning." Review of Financial Studies 33(5), 2223–2273. https://academic.oup.com/rfs/article/33/5/2223/5758276
• Gu, S., Kelly, B. & Xiu, D. (2021). "Autoencoder Asset Pricing Models." Journal of Econometrics 222(1), 429–450.
• Coles, J., Heath, D. & Ringgenberg, M. (2022). "On Index Investing." Journal of Financial Economics 145(3), 665–683. https://www.sciencedirect.com/science/article/abs/pii/S0304405X22001143
• Lopez-Lira, A. & Tang, Y. (2023). "Can ChatGPT Forecast Stock Price Movements? Return Predictability and Large Language Models." SSRN 4412788; arXiv:2304.07619. https://arxiv.org/abs/2304.07619

Books

• Lopez de Prado, M. (2018). Advances in Financial Machine Learning. Wiley.
• Lopez de Prado, M. (2020). Machine Learning for Asset Managers. Cambridge / CFA Institute Research Foundation.
• Lopez de Prado, M. (2023). Causal Factor Investing. Cambridge.
• Lo, A. (2017). Adaptive Markets: Financial Evolution at the Speed of Thought. Princeton University Press. https://press.princeton.edu/books/paperback/9780691191362/adaptive-markets
• Hull, J. (2022). Options, Futures and Other Derivatives (11th ed.). Pearson.
• Tuckman, B. & Serrat, A. (2022). Fixed Income Securities: Tools for Today's Markets (4th ed.). Wiley.
• Shreve, S. (2004). Stochastic Calculus for Finance II: Continuous-Time Models. Springer.
• Cartea, Á., Jaimungal, S. & Penalva, J. (2015). Algorithmic and High-Frequency Trading. Cambridge.
• Grinold, R. & Kahn, R. (2000). Active Portfolio Management. McGraw-Hill.
• Ilmanen, A. (2022). Investing Amid Low Expected Returns. Wiley.
• Ang, A. (2014). Asset Management: A Systematic Approach to Factor Investing. Oxford.
• Zuckerman, G. (2019). The Man Who Solved the Market. Penguin.
• Lewis, M. (2014). Flash Boys: A Wall Street Revolt. Norton.
• Mallaby, S. (2010). More Money Than God: Hedge Funds and the Making of a New Elite. Penguin.
• Spitznagel, M. (2021). Safe Haven: Investing for Financial Storms. Wiley.
• Soros, G. (1987). The Alchemy of Finance. Wiley.

Regulator and central-bank sources

• Bank for International Settlements (BIS) Triennial Survey of FX and OTC derivatives.
• BIS Quarterly Review.
• Basel Committee on Banking Supervision — Fundamental Review of the Trading Book (FRTB) standard.
• US Securities and Exchange Commission — Reg NMS, Reg ATS, market-structure proposals 2022–2024.
• Federal Reserve Bank of Boston (2025), "Could the Growth of Private Credit Pose a Risk to Financial System Stability?" https://www.bostonfed.org/publications/current-policy-perspectives/2025/could-the-growth-of-private-credit-pose-a-risk-to-financial-system-stability.aspx
• FRBNY Liberty Street Economics blog.
• FDIC (2025). Risk Review. https://www.fdic.gov/analysis/2025-risk-review.pdf
• Office of Financial Research (OFR) annual reports and working papers.
• IMF Global Financial Stability Report.

Industry / practitioner research

• AQR Capital Management — research library, especially Asness, Frazzini, Pedersen, Israel, Ilmanen.
• Two Sigma Insights.
• Bridgewater — "Daily Observations" excerpts; "The All Weather Story" public note. https://www.bridgewater.com/research-and-insights/the-all-weather-story
• Cboe Global Markets — Volatility Insights blog. https://www.cboe.com/insights/
• BPI (Bank Policy Institute) — FRTB primers. https://bpi.com/why-is-the-frtb-expected-shortfall-calculation-designed-as-it-is/
• Newfound Research / Corey Hoffstein.
• Damodaran's NYU Stern site (cost of capital, equity-risk-premium data, valuation lectures).
• Numerix — 0DTE market commentary. https://www.numerix.com/resources/blog/zero-day-options-0dte-start-2025-bang

Press / commentary

• Bloomberg — Matt Levine's "Money Stuff" (newsletter and column), Joe Weisenthal & Tracy Alloway "Odd Lots."
• FT Alphaville; FT Markets section.
• Robin Wigglesworth's Trillions and FT coverage of index investing.
• Patrick O'Shaughnessy — Invest Like the Best podcast.
• John Hempton — Bronte Capital blog.
• Ben Carlson — A Wealth of Common Sense.

Crypto-specific

• Flashbots research and MEV-Boost documentation. https://boost.flashbots.net/
• Paradigm Research blog.
• a16z Crypto Research.
• Vitalik Buterin's blog (vitalik.eth.limo).
• Galaxy Research.
• Coin Metrics State of the Network.
• Uniswap V2 and V3 whitepapers.
• dYdX V4 documentation (docs.dydx.exchange).
• Hyperliquid documentation (hyperliquid.gitbook.io).
• 21Shares Research, "Perps explained: How Hyperliquid and dYdX are powering the next phase of crypto trading." https://www.21shares.com/en-eu/research/perps-explained-how-hyperliquid-and-dydx-are-powering-the-next-phase-of-crypto-trading

Topical secondary sources cited above

(In addition to all of the above.)

• Wikipedia entries used as overview anchors: Black-Scholes model, Fama-French three-factor model, Carhart four-factor model, Markowitz model, Modern portfolio theory, Adaptive market hypothesis, Grossman-Stiglitz paradox, Universa Investments, Mark Spitznagel, Glosten-Milgrom model, Purged cross-validation, Deflated Sharpe ratio.
• Speedrun Ethereum, "Impermanent Loss Explained." https://speedrunethereum.com/guides/impermanent-loss-math-explained
• Hummingbot, "Guide to the Avellaneda-Stoikov Strategy." https://hummingbot.org/blog/guide-to-the-avellaneda--stoikov-strategy/
• Various CFA Institute publications (Hill/Nadig/Hougan technical analysis review, etc.).

Glossary appendix

A working vocabulary of 180+ trading terms, in plain English.

Active management — investment approach that attempts to beat a benchmark via security selection or market timing. Contrast: passive.

Adverse selection — the cost a market maker pays for trading against participants who have information he doesn't.

Algorithmic trading — any trading strategy executed by computer. Often (incorrectly) used as synonym for HFT.

Alpha — risk-adjusted excess return; the intercept in a factor regression. The thing managers claim to deliver.

Alternative data — non-traditional data sources used in trading (satellite, web scraping, credit card, geolocation).

AMM (Automated Market Maker) — an on-chain trading mechanism in which liquidity is supplied to a smart contract that prices trades algorithmically against a bonding curve, rather than via an order book.

Annualized return — a return scaled to a one-year period. Annualization of a daily return: (1 + r_daily)^252 − 1.

Arbitrage — risk-free profit from price discrepancies. Pure arbitrage is rare; "statistical arbitrage" is the more common, risky kind.

Ask price — the price at which a market maker will sell.

At-the-money (ATM) — strike equals current spot.

Backtesting — simulating a strategy on historical data.

Backwardation — when futures trade below spot. Earns positive roll yield for longs.

Basis — the difference between two related prices: futures vs. spot, cash bond vs. CDS, etc.

Basis trade — long one of two related instruments, short the other, capturing the spread.

Bear market — a market in extended decline (often defined as >20% drawdown).

Beta — sensitivity of an asset's return to the market's return.

Bid-ask spread — difference between the best bid and best ask.

Black swan — a rare, high-impact event outside the model's distribution.

Block trade — a single large transaction, often negotiated off-exchange.

Bond — debt security paying periodic coupons and returning principal at maturity.

Book-to-market — accounting book value divided by market capitalization. High = "value"; low = "growth."

Breadth (Grinold) — the number of independent active bets a manager makes per unit time.

Bull market — extended upward trend.

Cap-weighted — index weights proportional to market capitalization. The default for most equity indices.

Carry — the return you earn if nothing changes.

CDS (Credit Default Swap) — insurance against bond default.

CEX — centralized cryptocurrency exchange (Binance, Coinbase, etc.).

Coin — primary native asset of a blockchain (BTC, ETH, SOL).

Cointegration — statistical relationship between two non-stationary series whose linear combination is stationary.

Concentrated liquidity — Uniswap V3-style provision of liquidity over a price range rather than across the full curve.

Contango — futures trade above spot. Negative roll yield for longs.

Convexity — second-derivative sensitivity (e.g., bond price to yield); benefits long-bond holders.

Counterparty risk — the risk that the other side of a trade fails to deliver.

Covenant — a condition in a debt agreement constraining the borrower.

CTA (Commodity Trading Advisor) — a hedge-fund category, predominantly trend-followers in futures.

Dark pool — an ATS that doesn't display quotes pre-trade.

Delta — option sensitivity to spot. Used as the primary hedge ratio.

Delta-neutral — a portfolio with zero net delta exposure.

Derivative — a security whose value derives from an underlying asset.

DEX — decentralized cryptocurrency exchange.

Discretionary — trading driven by human judgment rather than rules.

Dispersion trade — short index vol, long single-name vol. Bets that realized correlation < implied.

Drawdown — peak-to-trough decline.

ETF (Exchange-Traded Fund) — a fund whose shares trade intraday like a stock.

Efficient frontier — the set of portfolios with maximum expected return for each level of risk.

EMH (Efficient Market Hypothesis) — Fama's claim that prices reflect available information.

ESG — environmental, social, governance investing criteria.

Expected Shortfall (ES, CVaR) — average loss given a tail event.

Factor — a return-generating systematic risk (market, size, value, momentum, etc.).

Factor zoo — derisive label for the proliferation of published factors.

Fat tail — a probability distribution with more extreme outcomes than the normal distribution.

Flash crash — sudden, large, intraday price plunge that quickly reverses (e.g., May 6, 2010).

Floor (option) — minimum guaranteed price.

Funding rate — periodic payment between longs and shorts in a perpetual futures contract.

Gamma — option sensitivity to changes in delta; second derivative w.r.t. spot.

Greek (options) — partial derivative of option price w.r.t. an input variable.

Hedge — a position taken to offset risk in another position.

HFT (High-Frequency Trading) — short-horizon (sub-second) trading with substantial infrastructure investment.

Hurst exponent — a measure of long-range memory in time series; H > 0.5 suggests trending behavior.

Iceberg order — large limit order with most size hidden.

Implementation shortfall — the difference between the price at decision time and the realized average fill.

Implied volatility (IV) — the σ that makes the Black-Scholes model price match the observed market price.

Impermanent loss (IL) — the loss an AMM LP suffers vs. just holding the assets, due to arbitrage.

Index fund — passive fund tracking a benchmark index.

Information ratio (IR) — active return / tracking error.

Insider trading — illegal trading on material non-public information.

Institutional investor — non-retail: pension funds, endowments, sovereign wealth funds, mutual funds, hedge funds, insurance companies.

In-the-money (ITM) — call with strike < spot; put with strike > spot.

Inverse ETF — ETF that delivers the inverse of an index's daily return (often levered).

JIT (Just-In-Time) liquidity — Uniswap V3 strategy of placing narrow ranges around pending trades.

Kelly criterion — formula for optimal bet sizing given known edge and odds; rarely used at full Kelly because of tail risk.

Kurtosis — fourth moment of a distribution; measures tail fatness.

Latency arbitrage — exploiting price discrepancies across venues using infrastructure-speed advantages.

Leverage — borrowing to amplify returns (and losses).

Limit order — order to buy/sell at a specified price or better.

Liquidity — ease of trading without moving the price.

Liquidity provider (LP) — entity that supplies the resting orders or AMM reserves that takers trade against.

Long-only — strategy restricted to buying (no short selling).

Long-short — strategy that simultaneously holds long and short positions.

Macro — investing/trading based on macroeconomic views (rates, currencies, commodities, indices).

Margin — collateral required to support a leveraged position.

Market maker — participant who quotes both bid and ask, earning the spread.

Market order — order to execute immediately at the best available price.

Mark-to-market (MTM) — valuing a position at current market prices.

Maximum drawdown (MDD) — largest peak-to-trough loss.

Mean reversion — tendency for prices to return to a long-run average.

Mean-variance optimization — Markowitz portfolio construction balancing expected return and variance.

Meta-labeling — Lopez de Prado technique: secondary ML model decides whether to act on a primary signal.

MEV (Maximal Extractable Value) — value extractable by reordering, inserting, or censoring transactions on a blockchain.

Microstructure — the high-resolution mechanics of how prices form (order books, queue position, market impact).

Momentum — tendency of recent winners to continue winning over months.

Moneyness — how in/out-of-the-money an option is.

Naked option — sold option without an offsetting hedge.

NBBO (National Best Bid and Offer) — best displayed prices across US equity venues.

OAS (Option-Adjusted Spread) — credit spread adjusted for embedded optionality.

OCO order — One-Cancels-Other.

OMS (Order Management System) — software for routing and tracking trades.

Open interest — total number of outstanding futures/options contracts.

Order book — list of resting limit orders.

Out-of-the-money (OTM) — call with strike > spot; put with strike < spot.

Overfitting — fitting noise rather than signal in backtests.

P&L (Profit and Loss) — trader's daily/cumulative gain or loss.

Pairs trading — long one stock, short a related stock; bet on spread mean-reversion.

Passive investing — owning a market-cap-weighted basket, no active selection.

Payoff diagram — visualization of an option strategy's terminal value vs. underlying price.

PBS (Proposer-Builder Separation) — Ethereum design where validators outsource block construction to specialized builders.

Perp / perpetual swap — futures contract without expiry, kept tethered to spot via funding payments.

PFOF (Payment For Order Flow) — payment from wholesaler to retail broker for routed order flow.

Position sizing — choosing trade size as fraction of capital.

Premium — option price; also the spread of a perpetual over spot.

Price impact — how much a trade moves the market.

Private credit — non-bank lending to corporate borrowers, usually held to maturity.

Purged k-fold CV — Lopez de Prado time-series-safe cross-validation removing label-overlap leakage.

Put-call parity — algebraic identity relating call, put, spot, and discounted strike.

Quant — systematic trader using mathematical/statistical models.

Quote stuffing — flooding venues with rapid quote changes; arguably manipulative.

Reflexivity — Soros's idea that prices and fundamentals influence each other.

Reg NMS — US 2005 regulation establishing the NBBO and trade-through rule.

Repo — repurchase agreement; short-term collateralized borrowing.

Reservation price — Avellaneda-Stoikov inventory-adjusted mid.

Risk parity — portfolio construction equating risk contribution across asset classes.

Risk premium — expected return above the risk-free rate as compensation for bearing risk.

Roll yield — return from rolling a futures position to the next contract, positive in backwardation.

Sandwich attack — MEV strategy bracketing a victim trade with adversarial buys/sells.

Sharpe ratio — risk-adjusted return measure.

Short selling — borrowing a security to sell, planning to buy back later at lower price.

Skew — third moment; also the slope of implied vol across strikes.

Slippage — gap between expected and realized execution price.

SOR (Smart Order Router) — software that splits orders across venues to optimize execution.

Sortino ratio — Sharpe variant that penalizes only downside vol.

Spread — bid-ask difference; or yield difference between two bonds.

Stablecoin — crypto token pegged to a fiat currency.

Stat arb — statistical arbitrage; market-neutral mean-reversion at scale.

Stop-loss — order placed to limit loss on a position.

Strangle — long OTM call + long OTM put.

Straddle — long ATM call + long ATM put.

Survivorship bias — bias from analyzing only surviving entities; especially insidious in backtests.

SWAP — over-the-counter derivative exchanging cash flows.

Synthetic — replicating a payoff via combinations of other instruments.

Tail risk — risk of extreme adverse outcomes.

Tail hedge — long-vol/long-OTM-put strategy designed to pay off in crashes.

Theta — option time decay.

Tick — minimum price increment.

TIF (Time-in-Force) — instruction governing how long an order is valid (DAY, GTC, IOC, FOK).

Tracking error — std of (portfolio − benchmark) returns.

Trade-through — executing at a price worse than the NBBO; forbidden by Reg NMS.

Trend-following — strategy buying recent winners, selling recent losers.

TWAP (Time-Weighted Average Price) — execution benchmark spreading trades evenly through a period.

UMD ("Up Minus Down") — Carhart's momentum factor.

Unhedged — exposed to underlying price moves.

Universe — set of securities a strategy considers.

Value at Risk (VaR) — quantile loss measure.

Vanilla — standard / unexotic instrument.

Vega — option sensitivity to volatility.

Volatility (σ) — annualized standard deviation of returns.

Vol risk premium (VRP) — average gap between implied and realized volatility.

VWAP (Volume-Weighted Average Price) — execution benchmark matching the day's volume profile.

Wash trade — fake trade between coordinated accounts to create false volume; manipulative.

Whale — large holder of a crypto asset, with market-moving influence.

Wholesaler — entity that internalizes retail orders (Citadel Securities, Virtu, G1 Execution Services, Two Sigma Securities).

X-asset arbitrage — strategy arbitraging across asset classes.

Yield curve — plot of yield vs. maturity for a class of bonds.

Z-score — deviation from mean in std units; used in mean-reversion signals.

Z-spread — parallel shift to the zero curve that prices a bond at its market price.

0DTE — option expiring same day as traded.

1/N — naive equal-weight portfolio. Surprisingly competitive vs. optimized portfolios in many studies (DeMiguel-Garlappi-Uppal 2009).

2-and-20 — traditional hedge-fund fee: 2% management + 20% performance.

60/40 — classic balanced portfolio: 60% stocks, 40% bonds.

End of report. Word count: approximately 28,000.