Build an auditable distribution from cash yield, growth, valuation, dilution, inflation, and implementation—then test whether the decision survives plausible errors.
One annual percentage hides the model and its uncertainty
Expected return is the mean of a conditional distribution under stated assumptions, not the return promised in each year. Declare horizon, currency, nominal or real convention, arithmetic or geometric mean, reinvestment, cash flows, benchmark, and uncertainty. Arithmetic average return describes a one-period expectation; compound annual growth describes the realized path between beginning and ending wealth and is generally lower when returns vary. Real return should be calculated as (1 + nominal return) / (1 + inflation) − 1. For example, 8% nominal with 3% inflation is about 4.85% real, not exactly 5%. A U.S. stock history from 1926 through 2023 can provide context, but it ends before the current date, represents one successful country, and blends changing valuations, inflation, constituents, and regimes. [1][8] It is not a neutral forecast for the next decade or for a concentrated strategy. Show a range, downside path, sequence sensitivity, and which assumptions would change the decision. Without those fields, a single percentage is marketing precision.
Table 1. Return conventions| Field | Required definition | Failure |
|---|
| Horizon | Dates and cash flows | Annual number without period |
| Mean | Arithmetic or geometric | Mixed conventions |
| Inflation | Index and timing | Simple subtraction |
| Currency | Base and hedge | Hidden FX |
| Uncertainty | Range and path | Point estimate |
Convention
A return estimate needs horizon, mean, currency, inflation, and cash-flow rules.
Historical baselines require point-in-time universes and predefined regimes
Construct historical evidence with the universe that existed on each date, delistings and distributions, executable rules, and costs. Current constituents, surviving funds, or a sample selected after observing winners inflate the baseline. Store source, version, code, corporate-action treatment, calendar, currency, and missing-data policy. Segment regimes only with definitions fixed before evaluating returns; otherwise favorable labels become another form of selection. Regime analysis describes conditional sensitivity but cannot know which state will occur next or justify giving the preferred state a larger probability. Compare countries, subperiods, inflation states, rates, volatility, and the contribution of a few exceptional years. Keep the same strategy definition in every segment and report effective sample sizes. Point-in-time validation is essential, but a clean backtest remains an estimate subject to model and structural uncertainty. [5] If failures, costs, or constituent history are unavailable, do not repair the evidence with an invented conservative haircut; mark the baseline insufficient and widen or suspend the conclusion.
Table 2. Historical baseline controls| Question | Required evidence | Red flag |
|---|
| Universe | Point-in-time and delistings | Survivors |
| Return | Total and net | Price-only gross |
| Regime | Predefined | Labeled afterward |
| Reproduction | Source, version, code | Black box |
History
A winning country and surviving funds are not a neutral forecast.
Valuation bounds scenarios but does not time returns
For a stock or market, model cash yield, per-share fundamental growth, valuation change, dilution or net repurchase, and their interactions. Starting CAPE or another multiple has had a noisy relationship with later long-horizon returns, but it neither forecasts the next quarter nor guarantees mean reversion. [2] Define normalized earnings, margins, cyclicality, inflation exposure, payout, and share count. A stock bought at 30 times earnings can earn through per-share growth and cash distributions even if its multiple does not expand; compression reduces the result and expansion increases it. Worked example: price $100, EPS $5, starting P/E 20, 5% annual EPS and dividend growth, and a $2 first-year dividend. After five years, EPS is about $6.38. At an exit P/E of 18, terminal price is about $114.86; five undiscounted dividends total about $11.05, so terminal wealth without reinvestment or tax is about $125.91, roughly 4.72% annualized. With P/E 20, the result is higher. The calculation is a sensitivity, not a valuation verdict.
Table 3. Valuation decomposition| Driver | Base input | Sensitivity | Caution |
|---|
| Per-share growth | Normalized path | Higher/lower | Margins and dilution |
| Cash yield | Payout path | Cut/growth | Reinvestment |
| Exit multiple | Defensible range | Compression/expansion | Not a timer |
| Inflation | Named index | Multiple paths | Real conversion |
Valuation
A long-horizon anchor is not an entry clock.
Base, bull, bear, and break cases must change the same drivers
Use base, favorable, adverse, and thesis-break cases that alter the same identities: unit or revenue growth, margins, share count, cash payout, terminal valuation, inflation, financing, and cost. The bull case need not be deliberately harder to reach than the bear case; both require evidence and internal consistency. A thesis-break case differs from a delayed-return case by making the economic mechanism false. Assign probabilities only from a calibrated method and show sensitivity, because a weighted mean can hide a ruinous tail and create false precision. Preserve correlations among drivers: higher growth can require investment and dilution; higher rates can affect both valuation and financing; weak liquidity can raise costs exactly in the adverse case. Compare the full distribution with a feasible alternative, liability, and portfolio role—not a hurdle selected to force a preferred answer. The decision tree is: validate identities, inspect survival in the adverse and break cases, test sensitivity, then decide whether more evidence, a different size, or no action is warranted. It never converts a scenario midpoint into an order.
Table 4. Coherent scenarios| Case | Drivers | Output | Decision use |
|---|
| Base | Central ranges | Distribution | Compare alternative |
| Favorable | Consistent upside | Upper range | Plausibility |
| Adverse | Consistent downside | Lower/tail | Capacity |
| Break | Mechanism false | Loss path | Invalidation |
Scenarios
Bull and bear cases must preserve the same economic identities.
Fees, tax, slippage, and capacity enter before the decision
Calculate gross, pre-tax net, and after-tax scenarios separately. Expense ratios follow the prospectus and may already be embedded in reported fund returns; subtracting them again is double counting. Spread, impact, and slippage depend on turnover, order size, liquidity, volatility, and execution; borrow and financing vary through time; tax depends on account, lot, holding period, income, jurisdiction, and current law. [3][4] Model capacity nonlinearly when participation rises. Do not invent a “behavioral drag” percentage: model specific rule breaks or treat adherence as a qualitative risk and measure it with live records. Use executable prices and scenario-specific costs so the bear case does not retain calm-market spreads. Reconcile cost components to avoid overlap among a fund return, fee, turnover model, and account statement. An expected return that exists only with perfect fills, unlimited liquidity, or a single tax rate is a laboratory output, not an investable estimate. The key point is that implementation uncertainty belongs inside the distribution, not in a footnote below its midpoint. Reconcile modeled turnover to actual orders, separate fixed and variable fees, and stress delayed or partial execution. For tax, distinguish income, realized short- and long-term gains, loss offsets, distributions, withholding, and account-specific deferral; use current law and show a no-tax comparison rather than pretending one effective rate applies to every path. If the estimate spans several investors or jurisdictions, publish separate scenarios. A wider honest range is more useful than a narrow net return built from averaged frictions that cannot occur together.
Table 5. Implementation costs| Cost | Source | Model | Error |
|---|
| Fee | Prospectus/account | Charge convention | Double count |
| Spread/impact | Quotes and volume | Turnover and size | Constant cost |
| Tax | Account and lots | Scenario | One rate |
| Behavior | Adherence records | Specific breach | Invented haircut |
Costs
A strategy that fails after executable costs has no usable expected return.
A 14% claim must be decomposed, not reduced by arbitrary haircuts
A claimed 14% expected return must be decomposed into market beta, factor exposures, selection, leverage, valuation change, cash yield, and implementation. Do not start with a historical 10% and apply fixed −1%, −2%, or −0.5% haircuts until the answer appears prudent. Estimate each driver as a range with source, horizon, and correlation; attribute factor returns consistently and reserve unproven alpha as uncertainty rather than a positive plug. Value and momentum evidence can inform a mechanism but does not guarantee future premiums. [6][7] Worked worksheet: beta contribution has a conditional range; factor contribution has held-out evidence and error; valuation has a terminal-multiple sensitivity; cash yield uses an explicit payout path; leverage includes financing and path risk; costs use turnover and capacity. Compound the scenario cash flows rather than adding incompatible annualized figures. The resulting distribution may include negative outcomes. If most of the 14% remains labeled “selection alpha” without independent evidence, the honest conclusion is that the headline claim is unsupported—not that its correct value is a narrower number chosen by judgment.
Table 6. 14% claim decomposition| Component | Evidence | Output | Uncertainty |
|---|
| Market beta | Conditional baseline | Range | Regime |
| Factor | Held-out attribution | Range | Decay |
| Selection | Independent validation | Range or unsupported | Sampling |
| Valuation/yield | Cash-flow model | Sensitivity | Terminal path |
| Costs | Executable implementation | Net range | Capacity |
Decomposition
Rebuild the 14% claim instead of subtracting arbitrary discounts.
Margin of safety means robustness to several simultaneous errors
Margin of safety in expected-return work is the stability of a decision when several plausible assumptions are wrong together. It is not automatically “cheap,” a fixed gap to a universal hurdle, or a rule to halve one heroic assumption and set two to zero. Define the alternative, liabilities, risk capacity, liquidity, and decision-specific required compensation. Then stress lower growth, margin compression, dilution, a worse exit multiple, higher costs, delayed cash flows, and unfavorable tax or inflation in coherent combinations. Record the break-even value of each driver and the smallest joint change that reverses the decision. A robust conclusion survives multiple reasonable parameter sets and remains implementable; a fragile one depends on a narrow point, even if its midpoint looks attractive. Conservatism also has opportunity cost, so report false-negative risk rather than assuming a lower estimate is always wiser. Update the model when evidence changes, not when price alone creates discomfort. Practical takeaway: audit the decomposition, distribution, and break-even assumptions; do not buy, sell, or size solely because a scenario clears a round percentage.
Robustness
Margin of safety is survival across plausible joint errors.
Related analysis
expected returnscenario analysisvaluationrisk controlportfolio planning
Sources & Further Reading
- Ibbotson Associates / Morningstar. Stocks, Bonds, Bills, and Inflation (SBBI) Yearbook data, 1926–2023.
- Robert J. Shiller, Online Data: U.S. Stock Market Data, CAPE and related series.
- U.S. Securities and Exchange Commission. Investor Bulletin: Mutual Fund Fees and Expenses. Source
- U.S. Securities and Exchange Commission. Understanding Your Brokerage Account Costs and Fees. Source
- CFA Institute. The Importance of Point-in-Time Data in Backtesting.
- Fama, Eugene F., and Kenneth R. French. “The Cross-Section of Expected Stock Returns.” Journal of Finance 47, no. 2 (1992): 427–465. Source
- Asness, Clifford S., Tobias J. Moskowitz, and Lasse H. Pedersen. “Value and Momentum Everywhere.” Journal of Finance 68, no. 3 (2013): 929–985. Source
- U.S. Bureau of Labor Statistics. Consumer Price Index data and inflation measures.