Starting from MPT's Separation Theorem (all investors hold the same Tangency Portfolio), CAPM derives an equilibrium pricing model. If all investors hold the market portfolio, then in equilibrium the market portfolio IS the Tangency Portfolio. The Security Market Line (SML) plots expected return against beta: E(Rᵢ) = Rƒ + βᵢ × (E(Rₘ) - Rƒ). Every asset and portfolio in equilibrium should lie on the SML — assets above the line offer positive alpha (expected return exceeds CAPM prediction), assets below offer negative alpha. The slope of the SML is the equity risk premium (ERP).

CAPM delivers three key testable predictions: (1) the market portfolio lies on the Efficient Frontier; (2) expected return is a linear function of beta; (3) beta is the only risk factor that matters (idiosyncratic risk is diversified away and therefore uncompensated). Testing these predictions against historical data has been the central project of empirical finance research for 60 years — and the results are mixed: beta does explain return differences broadly, but the relationship is flatter than CAPM predicts (low-beta stocks outperform their CAPM predictions, high-beta stocks underperform), and factors beyond beta (size, value, momentum) explain substantial return variation.

CAPM's most important practical application is estimating the cost of equity for corporate finance decisions. Weighted Average Cost of Capital (WACC) = (E/V) × Kₑ + (D/V) × Kd × (1-T), where Kₑ (cost of equity) is calculated using CAPM: Kₑ = Rƒ + β × ERP. For a company with beta 1.2, risk-free rate 4%, and ERP of 5%, Kₑ = 4% + 1.2 × 5% = 10%. This cost of equity is then used as the hurdle rate in capital budgeting (projects must earn at least the WACC to create shareholder value) and in DCF valuation (the discount rate applied to future cash flows).

The ERP (equity risk premium) is the most uncertain input in CAPM applications. Historical ERP (US equities vs. T-bills, 1928-2023) is approximately 5.5-6.5% depending on the period and methodology. Forward-looking ERP estimates (implied by current valuations) are typically 3.5-5%. The choice of ERP substantially affects valuation — a 1 percentage point higher ERP increases the cost of equity by beta points and reduces discounted valuations meaningfully. Practitioners disagree on whether to use historical or implied ERP; many use a range (4-6%) and assess valuation sensitivity.

The empirical failures of CAPM are well-documented. Beta alone has virtually no power to explain the cross-section of stock returns in Fama-French's research — a finding that directly motivated the multi-factor models. The low-volatility anomaly (low-beta stocks earn higher risk-adjusted returns than CAPM predicts) is one of the most robust findings against CAPM's prediction that return should be linearly increasing in beta. Value and size premiums are unexplained by CAPM.

These failures have spawned CAPM extensions: the Fama-French three-factor model adds size and value; the five-factor model adds profitability and investment patterns; the Carhart four-factor model adds momentum. Each extension improves explanatory power over single-factor CAPM. Yet CAPM remains the standard for cost-of-capital estimation in corporate finance because its simplicity and interpretability outweigh its empirical imperfections for practical applications — a 'good enough' model is more useful than a theoretically superior but complex one when explaining investment decisions to boards and investors.