Modern Portfolio Theory: The Foundation and Its Blind Spots

MPT's central contribution is demonstrating mathematically that portfolio risk (variance) depends not just on the volatility of individual assets, but on the correlations between them. Two assets that individually have 20% volatility can be combined into a portfolio with less than 20% volatility if their returns are not perfectly correlated.

The efficient frontier is the set of portfolios that maximize expected return for a given level of risk, or minimize risk for a given level of expected return. All portfolios below and to the right of this frontier are suboptimal — you are either taking too much risk for the return you are receiving, or getting too little return for the risk you are taking.

MPT is built on five assumptions that hold reasonably well in normal conditions and fail in crisis conditions.

First: returns are normally distributed. Fat tails violate this assumption precisely when it matters most.

Second: correlations are stable over time. In reality, correlations increase dramatically during crises — the diversification benefits MPT predicts evaporate exactly when you need them.

Third: investors can borrow and lend at the risk-free rate. Leverage availability and cost vary dramatically across market regimes.

Fourth: all investors have identical time horizons. In practice, liquidity needs and drawdown tolerance vary enormously.

Fifth: markets are frictionless. Transaction costs, liquidity constraints, and bid-ask spreads can substantially alter real-world performance.

Even accepting MPT's framework, the optimization process requires inputs — expected returns, volatilities, and correlations — that must be estimated from historical data. Small errors compound dramatically through the optimization process, producing portfolios that appear "optimal" but are extremely sensitive to estimation assumptions.

This phenomenon, sometimes called the "error maximizer" problem, means that mean-variance optimized portfolios frequently concentrate in a small number of assets with high estimated returns and low estimated correlation — a precision that is statistically unjustified given the noise in return estimation. Robustness matters more than precision in practice.

Despite its limitations, MPT's core insights remain foundational. Diversification genuinely reduces risk in normal market conditions. The distinction between market risk (systematic, compensated) and idiosyncratic risk (specific to a single asset, uncompensated) remains the basis for understanding why broad equity exposure makes more sense than concentrated single-stock bets.

The Capital Asset Pricing Model (CAPM), which builds on MPT, correctly identifies that investors should not expect compensation for risks they could have diversified away. These are durable insights that inform sound portfolio construction even when the specific optimization outputs are unreliable.

Where MPT models what portfolios should do on average under normal conditions, stress testing models what they actually do under extreme conditions. The two approaches answer different questions.

MPT asks: given expected returns and historical average correlations, what is the theoretically optimal allocation? Stress testing asks: given the actual correlation structure that prevailed during the 2008 crisis, the 2020 COVID crash, or the 2022 rate shock, what did this portfolio actually experience? The combination — theory for structure, scenario analysis for validation — produces more robust portfolios than either approach alone.