Capital market expectations — sometimes called capital market assumptions — are the quantitative forecasts of returns, risks, and correlations that drive every serious portfolio construction decision. Once you have defined your objectives and constraints in an investment policy statement, the next step is translating those goals into a portfolio. That translation depends entirely on the quality of your capital market expectations. This guide covers the systematic framework professionals use to set these expectations, from historical analysis and DCF models to equilibrium pricing and expert judgment.
What Are Capital Market Expectations?
Capital market expectations (CME) are an investor’s forecasted expected returns, standard deviations, and correlations for each asset class under consideration. They are the required inputs for any formal asset allocation process — without them, optimization is impossible.
Capital market expectations are strategic, horizon-specific forecasts for asset classes — not short-term trading signals. A typical CME exercise produces expected returns, volatilities, and correlations for 5-10+ asset classes over a 5- to 10-year horizon, forming the foundation for mean-variance optimization and strategic asset allocation.
CME focuses on the macro level — forecasts for entire asset classes like U.S. equities, investment-grade bonds, or emerging market debt — not individual securities. The scope is substantial: with just 8 asset classes, an analyst needs 8 expected returns, 8 standard deviations, and 28 pairwise correlations — 44 distinct estimates. As the number of asset classes grows, the correlation matrix expands rapidly, making disciplined estimation methods essential.
How to Set Capital Market Expectations: The 7-Step Framework
The CFA Institute’s framework for setting capital market expectations provides a structured, repeatable process. Following these steps ensures consistency and accountability:
- Specify the expectations needed — Identify which asset classes, time horizons, and return measures (nominal vs. real, pre-tax vs. after-tax) the investment process requires.
- Research the historical record — Study the return drivers, risk characteristics, and long-term performance of each asset class. Understand not just what happened, but why.
- Specify the methods and models — Select and justify your approach: historical extrapolation, DCF models, equilibrium models, surveys, or a combination.
- Determine the best data sources — Evaluate data quality, investability of indices, free-float corrections, and whether data frequency matches your time horizon.
- Interpret the current investment environment — Apply judgment to reconcile conflicting signals. Ensure all assumptions are mutually consistent across asset classes.
- Provide and document the expectations — Present the final set of returns, volatilities, and correlations with clear documentation of assumptions and reasoning.
- Monitor actual outcomes and provide feedback — Compare realized results to forecasts. Assess whether errors were due to bad luck or bad process, and improve iteratively.
Steps 6 and 7 — documentation and feedback — are the most frequently skipped in practice, yet they are what separate institutional-quality CME from ad hoc guessing. Recording your assumptions creates accountability and allows you to learn systematically from forecast errors over time.
Historical Data Analysis: Challenges and Pitfalls
Historical data is the starting point for most CME exercises, but using it responsibly requires understanding its limitations. The Dimson, Marsh, and Staunton (2006) study of 17 major markets from 1900 to 2005 illustrates both the value and the challenge: real equity returns ranged from 4.6% (Belgium) to 10.1% (Sweden), while real bond returns ranged from -0.4% (Italy) to 3.7% (Denmark). The spread across countries is wide enough that choosing the “wrong” reference market can materially distort expectations.
Six common problems plague historical data analysis:
- Survivorship bias — Indices reflect only companies (or markets) that survived. Failed firms disappear from the data, overstating historical returns.
- Data mining bias — Repeatedly searching for patterns will inevitably find statistically significant but economically meaningless relationships. Always require an economic rationale before trusting a historical pattern.
- Time-period bias — Results are sensitive to start and end dates. The U.S. small-cap premium, for example, was concentrated almost entirely in 1975-1983; excluding that 9-year window, large caps outperformed over 1926-2001.
- Look-ahead bias — Using information that was not available to investors at the time the data was generated.
- Transcription errors — Data entry mistakes in long time series can be systematic rather than random.
- Appraisal (smoothed) data — Returns for illiquid assets like real estate and private equity are often based on appraisals rather than market transactions, which can significantly understate true volatility and bias correlations toward zero.
Appraisal-based data for illiquid assets can significantly understate true volatility and make risk-adjusted performance appear artificially attractive. If fed directly into an optimizer, smoothed data will cause dangerously high allocations to these asset classes. Always adjust for smoothing before using alternative asset data in a CME framework.
A fundamental trade-off governs every historical analysis: longer time series provide more statistical precision but may include irrelevant regimes, while shorter series are more relevant but noisier. The practical approach is to test for structural breaks and use only the relevant subsample. For a deeper look at how historical data informs equity risk premium estimates, see our dedicated article.
DCF Models for Expected Returns
Discounted cash flow models anchor expected returns to current market fundamentals rather than historical averages. They are forward-looking by construction, which makes them especially valuable for long-horizon capital market expectations.
Gordon Growth Model
Where:
- D1 / P0 — forward dividend yield (next year’s expected dividend divided by current price)
- g — expected long-term dividend (or earnings) growth rate, often proxied by nominal GDP growth
The Gordon model is simple and intuitive but assumes constant growth forever — a strong assumption that limits its accuracy for volatile or rapidly changing markets.
The Grinold-Kroner Model
The Grinold-Kroner model extends the Gordon model by explicitly accounting for share repurchases and changes in valuation multiples:
Where:
- D/P — expected dividend yield
- ΔS — expected change in shares outstanding (negative when companies buy back shares, so subtracting a negative adds to returns)
- i — expected inflation rate
- g — expected real earnings growth rate
- ΔPE — expected per-period percentage change in the price-to-earnings multiple
The three components are: income return (D/P − ΔS), nominal earnings growth (i + g), and repricing return (ΔPE).
| Component | Estimate |
|---|---|
| Dividend yield (D/P) | 1.75% |
| Repurchase yield (−ΔS) | +0.50% |
| Expected inflation (i) | 2.50% |
| Real earnings growth (g) | 3.50% |
| P/E repricing (ΔPE) | −0.75% |
| Total Expected Return | 7.50% |
Subtracting the 10-year Treasury yield of 5.0% produced a forward-looking equity risk premium estimate of 2.5% — notably lower than the long-run historical average of roughly 5%.
The Grinold-Kroner model also decomposes historical returns. The S&P 500’s 10.7% compound annual return from 1926 to 2001 broke down as: 4.4% income return + 4.8% nominal earnings growth + 1.5% repricing. Nearly all of the repricing component came from P/E expansion in the final two decades — a powerful reminder that valuation changes are the most volatile and least predictable component.
Fixed-Income Expected Returns: YTM Decomposition
For bonds, the yield to maturity provides a natural starting point for expected return estimation. The YTM can be decomposed into its economic building blocks:
Government bond yields embed the real risk-free rate, an inflation premium, and a maturity (term) premium for bearing duration risk. Corporate and non-sovereign bonds add default risk, liquidity, and potentially tax-related premia. Importantly, government duration bonds and credit-sensitive bonds behave very differently across economic cycles — a distinction that matters when mapping business cycle phases to fixed-income expectations.
Forecasting Covariances, Correlations, and Volatility
Expected returns receive most of the attention, but the covariance matrix — volatilities and correlations — is equally important and often harder to estimate well. With many asset classes, sample covariance matrices are extremely noisy and small estimation errors get amplified dramatically during optimization.
A shrinkage estimator blends the sample covariance matrix with a more stable structured target — such as a factor-model covariance matrix or one that assumes all pairwise correlations equal the average correlation. This is analogous to applying Bayesian skepticism: rather than trusting noisy sample data completely, you pull estimates toward a theoretically reasonable prior, reducing estimation error.
Shrinkage is particularly important when the number of asset classes is large relative to the number of historical observations. Under standard shrinkage-efficiency conditions with appropriately chosen weights, a structured target has been shown to improve on the raw sample estimate. Shrinkage estimators are critical for producing stable inputs for mean-variance optimization, where noisy covariance inputs can produce extreme and unstable portfolio weights.
Beyond shrinkage, analysts should account for two additional realities. First, correlations tend to rise sharply toward 1 during market crises — precisely when diversification is most needed. Building this asymmetry into stress-test scenarios is essential. Second, ARCH and GARCH models provide conditional volatility forecasts that reflect current market conditions by giving more weight to recent observations, capturing the well-documented tendency for volatility to cluster (large price swings followed by large swings).
The Singer-Terhaar Model: Equilibrium Risk Premiums
The Singer-Terhaar (1997) approach provides an equilibrium method for estimating asset class risk premiums. It is grounded in the International Capital Asset Pricing Model (ICAPM) but addresses a critical real-world imperfection that the standard model ignores: markets are not perfectly integrated.
Integrated Market Risk Premium
In a perfectly integrated global market, an asset’s risk premium depends on its systematic risk relative to the global investable market (GIM):
Only the portion of an asset’s risk that is correlated with the global market is priced — the term (RPM / σM) is the Sharpe ratio of the global market portfolio.
Segmented Market Risk Premium
In a completely segmented market, local investors price risk using only local factors. The asset’s correlation with the global market drops out (conceptually equals 1, since the local market is the reference portfolio):
Note that this simplified form assumes the local market Sharpe ratio equals the global market Sharpe ratio — a practical convenience that may not hold exactly for all markets. The segmented risk premium is always higher than the integrated premium because the correlation term (ρ ≤ 1) is absent.
Weighted Average for Partially Integrated Markets
Real markets are neither perfectly integrated nor completely segmented. Singer-Terhaar handles this by taking a weighted average:
Typical integration estimates from research: developed market equities and bonds ~80%, emerging market equities ~65%, U.S. real estate ~70%. Illiquidity premiums are added separately for asset classes where trading costs or lockup periods are material.
| Input | Value |
|---|---|
| Canadian equity volatility (σ) | 17% |
| Correlation with GIM (ρ) | 0.70 |
| Global market Sharpe ratio (RPM/σM) | 0.28 |
| Risk-free rate | 4.0% |
| Degree of integration (φ) | 80% |
Integrated RP = 17% × 0.70 × 0.28 = 3.33%
Segmented RP = 17% × 0.28 = 4.76%
Weighted RP = (0.80 × 3.33%) + (0.20 × 4.76%) = 2.66% + 0.95% = 3.62%
E(R) = 4.0% + 3.62% = 7.62%
The Black-Litterman model offers a complementary equilibrium approach that reverse-engineers expected returns from market-cap weights and blends them with investor views — the two methods triangulate well together.
Business Cycle Analysis as an Input to Capital Market Expectations
Business cycle phases are a key conditioning variable for capital market expectations. Rather than using unconditional long-run averages, professional investors adjust return and risk assumptions based on where the economy currently sits in the cycle. The output gap — defined as potential GDP minus actual GDP — is a central macro input: a positive output gap means the economy is operating below potential (slack, declining inflation), while a negative output gap means it is running above potential (overheating, rising inflation pressure).
| Phase | Equity CME | Fixed Income CME | Key Indicators |
|---|---|---|---|
| Initial Recovery | Above-average expected returns; cyclical and riskier assets outperform | Government yields bottoming; duration bonds still performing well | Business confidence rising, consumer confidence lagging, output gap large |
| Early Expansion | Positive but moderating returns; broad-based gains | Short rates rising as stimulus withdrawn; credit spreads tightening | GDP accelerating, unemployment falling, profits rising on lower unit costs |
| Late Expansion | Below-average expected returns; markets nervous | Yields rising across the curve; credit bonds outperforming duration | Output gap closing, inflation picking up, wage acceleration, tight monetary policy |
| Slowdown | Negative expected returns; lower-risk equity allocations warranted | Long government bonds rally sharply; yield curve may invert | Growth slowing despite still-rising inflation, leading indicators declining |
| Contraction | Elevated risk premiums; long-term expected returns rising as valuations compress | Aggressive monetary easing; government bonds outperform, credit spreads widen | GDP declining, unemployment rising, confidence collapsing, profits dropping |
Leading indicators — such as the yield curve slope, purchasing managers’ indices (PMI), and new unemployment claims — help identify which phase the economy is currently in. For a detailed framework on how business cycles inform sector-level allocation decisions, see our guide to sector rotation.
Survey and Judgment-Based Approaches
Surveys of economists and investment professionals provide a useful cross-check on model-based expectations. The Welch surveys of finance professors illustrate how consensus can shift: the median 30-year U.S. equity risk premium forecast fell from 7.0% in 1998 to 5.0% in 2001. Economic analysis — examining savings rates, productivity trends, demographic shifts, and government structural policies — provides the longest-horizon perspective on expected returns and complements shorter-term model outputs.
No single capital market expectations method dominates in all environments. Professional investors typically triangulate across historical analysis, DCF models, equilibrium models, and surveys — then apply judgment to reconcile conflicting signals. When the Grinold-Kroner model suggests 7.5% equity returns while surveys suggest 10%, investigating why they differ often reveals more insight than either number alone.
Methods for Setting Capital Market Expectations: Singer-Terhaar vs DCF vs Historical
Each of the three core CME methods has distinct strengths. The choice of method often depends on which asset class you are forecasting and what data is available.
Equilibrium Models (Singer-Terhaar)
- Theoretically grounded in the ICAPM
- Accounts for market integration and segmentation
- Produces consistent risk premiums across all asset classes simultaneously
- Requires subjective integration degree estimates
- Best for: global multi-asset portfolios — equities, bonds, real estate, commodities across developed and emerging markets
DCF Models (Grinold-Kroner / Gordon)
- Forward-looking, anchored to current market data
- Does not rely on historical average returns
- Sensitive to growth and discount rate assumptions
- Assumes stable dividend/earnings relationships
- Best for: equity and investment-grade bond return forecasts where current yields and growth rates are observable
Historical Analysis
- Data-rich with the longest available track record
- Easy to understand and communicate to stakeholders
- Backward-looking; may not reflect future regimes
- Subject to survivorship, time-period, and data mining biases
- Best for: calibrating volatility and correlation estimates, and as a sanity check against model-based return forecasts
In practice, institutional investors rarely rely on a single method. The best approach triangulates across all three and documents the reasoning behind the final expectations. How much you trust historical returns depends partly on your view of market efficiency — whether past pricing patterns persist or are quickly arbitraged away.
Common Mistakes When Setting Capital Market Expectations
Even experienced analysts fall into predictable traps when setting capital market expectations:
1. Relying solely on historical averages — Past returns reflect a single path through many possible outcomes. The U.S. was the most successful equity market of the 20th century — extrapolating its realized premium globally overstates expected returns.
2. Ignoring structural breaks — Tax regimes, monetary policy frameworks, and demographic shifts can permanently alter return distributions. Using pre-1980 inflation data to forecast returns in a post-inflation-targeting world is a classic error.
3. Data mining without economic rationale — Finding that a variable “predicted” returns historically is meaningless without a plausible economic mechanism. Always require out-of-sample testing and economic logic.
4. Anchoring to recent performance — Investors systematically overweight the most recent 3-5 years. After a bull market, return expectations rise; after a crash, they fall — precisely the opposite of what valuations typically imply.
5. Confusing statistical significance with economic significance — A correlation of 0.15 may be statistically significant with 100 years of data but economically meaningless for portfolio construction purposes.
6. Using unadjusted appraisal data for illiquid assets — Appraisal-based returns for real estate and private equity smooth true volatility, overstating risk-adjusted performance and causing optimizers to over-allocate to these asset classes.
7. Mixing inconsistent inputs — Combining nominal equity return forecasts with real bond yields, or mixing returns denominated in different currencies, produces internally contradictory expectations. All inputs must share the same return basis (nominal or real), currency, and time horizon.
Limitations of Capital Market Expectations
Even the most disciplined CME framework produces estimates, not certainties. Understanding these limitations is essential for using expectations appropriately:
All models are wrong — some are useful. The value of a disciplined capital market expectations process lies not in eliminating forecast error but in making errors smaller, more symmetric, and more transparent than they would be without a framework.
- Long-run expectations may not apply to short horizons — A 10-year return forecast tells you little about the next quarter. Conditional forecasts that account for current valuations and the business cycle are more useful for shorter periods.
- Regime changes invalidate historical relationships — Correlations that held for decades can break down during financial crises, precisely when stability matters most.
- Covariance instability — Correlations tend to rise sharply toward 1 during market stress, undermining the diversification benefits that the original expectations implied.
- Behavioral factors introduce systematic errors — Herding, overconfidence, and anchoring affect not just individual investors but the professional forecasting community as well.
- Model risk — Different reasonable assumptions produce materially different expected returns. The Grinold-Kroner model and Singer-Terhaar can disagree by 2-3 percentage points for the same asset class, even when both are correctly applied.
Frequently Asked Questions
Disclaimer
This article is for educational and informational purposes only and does not constitute investment advice. Capital market expectations examples cited are based on historical data and published research; actual future returns may differ materially. Always conduct your own research and consult a qualified financial advisor before making investment decisions.
