A topic in the Open Knowledge Graph — a free, open map of 15,290 topics and the order to learn them in.
Arbitrage Pricing Theory (APT), developed by Stephen Ross, generalizes CAPM by allowing multiple systematic risk factors to drive expected returns: E[rᵢ] = rₓ + β₁λ₁ + β₂λ₂ + … + βₖλₖ, where each βⱼ is the factor loading and λⱼ is the associated risk premium. APT is derived purely from the no-arbitrage condition — if expected returns were not proportional to factor exposures, investors could construct a zero-investment, zero-risk portfolio with positive expected return, which cannot persist. Empirical implementations include the Fama-French three-factor model (market, size, value) and the Carhart four-factor model (adding momentum), which substantially outperform one-factor CAPM in explaining cross-sectional return variation.
Understand the no-arbitrage derivation: if a diversified portfolio with zero factor exposure has positive expected return, it is an arbitrage opportunity that rational investors immediately exploit. Estimate a three-factor regression for a mutual fund to decompose its performance into factor exposures and true alpha.
CAPM, which you already know, makes a bold claim: a single factor — the return of the market portfolio relative to the risk-free rate — fully explains why different assets earn different expected returns. Every asset's expected return is determined entirely by its beta with the market. APT begins by asking: what if the economy has more than one source of systematic risk that investors care about and cannot diversify away? Maybe interest rate surprises, inflation shocks, and industrial output shocks all independently move portfolios in ways beta alone cannot capture. The Arbitrage Pricing Theory generalizes CAPM by allowing any number of such factors, each with its own premium.
The derivation is elegant and requires no utility theory or assumptions about investor preferences beyond risk aversion. It rests entirely on the no-arbitrage condition: if a portfolio can be constructed that has zero factor exposure (no systematic risk), requires zero net investment, and still earns a positive expected return, rational investors would demand infinite amounts of it, which is impossible in equilibrium. Therefore, in any well-functioning market, every unit of systematic risk exposure must be compensated by a commensurate expected return premium. The formula E[rᵢ] = rₓ + β₁λ₁ + β₂λ₂ + … + βₖλₖ follows directly: each factor loading βⱼ measures sensitivity to factor j, and each risk premium λⱼ is the market price of that risk.
From your multiple regression background, recognizing this as a regression structure is natural. Estimating factor exposures is exactly running a regression of asset returns on factor returns. The intercept — called alpha — measures return not explained by factor exposures: positive alpha means the asset earns more than its systematic risk profile warrants, which is either evidence of mispricing or evidence that your factor model is missing a relevant risk. This is how practitioners use the Fama-French three-factor model (market, SMB for size, HML for value) or the Carhart four-factor model (adding momentum): as a benchmark that strips out known systematic exposures so that true active management skill can be evaluated.
The key interpretive nuance is that APT itself does not tell you which factors to use. CAPM's single factor (the market) is theoretically motivated by equilibrium asset pricing — it is what all investors collectively hold. APT's factors are identified empirically by finding systematic co-movements in returns that command a premium. This flexibility is both APT's strength (it accommodates real-world complexity) and its weakness (it can be over-fitted by adding factors post-hoc until everything is explained). Good applied work requires theoretical motivation for each factor, not just statistical association.