From a company name to a source of dependence
A multi-industry portfolio may simultaneously depend on financing cost, issuer size and sensitivity to the economic cycle. Factor analysis searches for common components without replacing business analysis. Industry, leverage and statistical exposure are different dimensions and cannot be mechanically combined into one safety ranking.
Distinguish a characteristic from a return factor. Low P/B is a stock characteristic on a date. The return of a difference between model portfolios is the result of a specific portfolio-formation rule over a period. Calling one stock a “value factor” is insufficient: its result also depends on idiosyncratic events and market changes.
A portfolio difference is not an ordinary investment
In a fictional month, the return of the small-company group is 6% and the large-company group is 4%. The simple difference is 2 percentage points. But an investor who bought small companies did not necessarily earn a “2% premium” on all capital: its actual pre-cost return is 6%, while two points describe the comparison of two groups.
If in another month the first group loses 10% and the second loses 6%, the difference is −4 points. The existence of a size characteristic does not guarantee a positive addition in every period. Implementing a strategy, especially one with a short side, may require financing, collateral and costs absent from a simple research-series difference.
Regression does not prove causation
If a model explains a portfolio’s past return well, that means the selected factors fit the data over the chosen period. It does not establish one unique economic cause. Coefficients depend on observation window, portfolio composition and factor correlation. A short history and strategy changes make a stable conclusion especially difficult.
An unexplained residual cannot automatically be called manager skill. It may contain omitted risk, randomness, data errors and benchmark misspecification. Likewise, low measured exposure does not exclude a crisis relationship: the past window may not contain the relevant event. The SEC notes that historical performance does not predict future results.[27](#source-market-performance)
A premium as a requirement and as an observation
Separate required risk compensation in a valuation model, realised historical performance and an expectation of a future premium. They are different quantities. A high required return does not mean the investor will receive more money: uncertainty of outcome distinguishes a risky premium from a contractual payment. Discount-rate materials use risk estimates as assumptions, not promises.[93](#source-cv-rates)
When reading research, retain portfolio-formation rules, data-availability dates, costs and market restrictions. A Russian portfolio does not automatically inherit properties of an American research sample. Do not transfer historical figures across countries, currencies and periods without independent checking.
The result of factor analysis is a map of common dependencies and model limits. It can reveal a hidden bet on one mechanism among many names. It does not determine a personally suitable allocation, guarantee a positive premium or replace assessment of liquidity, access and possible permanent capital loss.
Do not count several labels as independent
A strategy may be called quality, value-oriented and profitability-focused at once. That does not prove three independent return sources. Selection rules may lead to the same companies or common industry exposures. Check actual holdings and formation method rather than a list of labels on a card.
Also distinguish a research factor from an accessible implementation. Short-sale limits, trade size, rebalance delays and costs can change the result. An owner of a product that follows only an approximate rule cannot be promised the return of the published series.
For an educational report, keep two tables: investment characteristics on the date and historical return dependencies over the window. The first describes composition, the second measured relationships. If conclusions differ, investigate the method rather than automatically declaring one table wrong. Neither table establishes a future premium.
**Related foundations:** [Growth and value: two languages of expectations, not two lists of winners](/guides/growth-and-value/); [Portfolio risk: losses, horizons and joint asset movements](/guides/portfolio-risk/).
Expected return: states first, premium second
An expectation is a probability-weighted sum of possible results, not the most pleasant forecast. MIT’s uncertainty lecture distinguishes expected monetary value from a person’s attitude to risk.[1](#source-rm-expected-utility) Set joint one-year states for two assets. All figures and probabilities are fictional and are used for reproducible arithmetic, not taken from current quotes.
| State after one year | Probability | Stock A: total return | Bond B: total return | Portfolio: 60% A and 40% B | | --- | --- | --- | --- | --- | | Downturn | 0.25 | −20% | 4% | −10.4% | | Normal year | 0.50 | 8% | 5% | 6.8% | | Upswing | 0.25 | 24% | 6% | 16.8% |
Probabilities sum to one. E(RA) = 0.25 × (−20%) + 0.50 × 8% + 0.25 × 24% = 5%; E(RB) = 5%. E(Rportfolio) = 0.25 × (−10.4%) + 0.50 × 6.8% + 0.25 × 16.8% = 5%. Checking by weights gives 0.60 × 5% + 0.40 × 5% = 5%. Linearity of expectation does not require independence. But loss distribution requires joint states: do not independently reshuffle A and B returns. Weights are beginning values; there is no intermediate rebalancing, contribution or cost.
Equal expectation does not make A and B equivalent. The portfolio’s negative-result probability in this table is 25%, and the worst stated state is −10.4%. This is not a limit on a real market loss: the table may omit a heavier outcome. Standard deviation under the stated probabilities is about 9.78 percentage points and describes only this model.
When expected return does not cover required compensation
Let a fictional one-year risk-free benchmark in the same currency be 3% and the model’s required portfolio return be 7%. Expected premium is 5% − 3% = 2 percentage points; required premium is 7% − 3% = 4 points. The missing two points are not a promised future bonus: they are a gap between expectation and requirement at the current price. Revisit price, scenarios or the conclusion, but do not raise the upswing probability to obtain a desired answer. The required-rate approach must be justified independently of this table.[2](#source-cv-rates)
At price 100, expected terminal value is 105. For the same future money and a 7% requirement, modelled initial value is 105 / 1.07 ≈ 98.13. This is not a market target: the payment distribution and risk were held constant. In reality, a price change can bring new information and different scenarios.
Test sensitivity to a rare outcome. If there is a 1% probability of total loss and a 99% probability of the prior distribution, expectation changes from 5% to 3.95%: 0.99 × 5% + 0.01 × (−100%). Such a loss need not appear in a short history, and its probability may be unknown. Historical average does not establish its absence. Do not attribute this illustration to a known investor’s strategy; it tests average sensitivity to an omitted tail.
CAPM, business risk and financing horizon are three different checks
In educational CAPM, ke = rf + β × ERP: rf is an agreed risk-free benchmark, ERP the required market premium and β sensitivity to the selected market. With fictional rf = 3%, ERP = 5% and β = 1.2, ke = 9%. This is a required model rate, not probability of loss, guaranteed return or mandatory future price growth.[1](#source-cv-rates) Market, window and observation frequency can change beta.
Beta describes joint movement, not a complete map of capital-loss risk. Small observed price fluctuations do not exclude customer loss, debt default, technological replacement or rare trading. An investor’s obligation to pay in six months also does not disappear because a business model spans ten years. Long horizon does not remove estimation error, financing risk or possible permanent loss. Business analysis, statistical exposure and a personal cash calendar supplement rather than replace one another.