Portfolio Risk: Exposure, Scenario Losses and Model Limits

From individual holdings to a combined loss

Portfolio risk concerns the ways a collection of positions can produce an unwanted outcome. It includes more than the volatility of each holding considered separately. Shared exposures can produce simultaneous losses, borrowing can magnify losses relative to equity, and assets with substantial quoted value may not produce usable cash when it is needed.

This lesson is a calculation workshop for a global audience. It builds on risk management, correlation and diversification and portfolio construction. Its specific job is to translate holdings into exposures and conditional loss estimates. The efficient frontier remains covered in Modern Portfolio Theory; no optimiser or preferred allocation is offered here.

Every portfolio, weight, return, correlation and cash deadline below is hypothetical. Dollar amounts are teaching units, not a recommendation for a reader's account. Examples exclude taxes and fees unless explicitly added. Sources were checked on 10 October 2026. This is financial education, not advice or a model portfolio to copy.

Decide what the risk question is

A portfolio can be described by several different questions: how much its marked value fluctuates, what it could lose in a specified scenario, how concentrated it is, or whether it can meet a payment by a deadline. A single number rarely answers all of them. A low estimate on one measure is not proof that every other risk is small.

For example, a value that changes infrequently might produce a smooth recorded return series. That smoothness could reflect how often the asset is valued rather than how easily it could be sold. Conversely, an actively traded asset can have frequent visible price changes while being comparatively straightforward to sell in small quantities under ordinary conditions.

The first analytical step is therefore a definition, not a calculation. Specify the portfolio boundary, the currency, the starting date, the horizon and the outcome being examined. “A 10% risk” is incomplete unless we know whether it means a scenario loss, an annualised volatility estimate, a probability or something else entirely.

The portfolio boundary also determines what the denominator includes. A securities account, all financial assets, or assets net of borrowing are different objects. A ratio computed for one should not be silently reported as a ratio for another. The examples begin with an unlevered $100,000 account and introduce borrowing separately.

Count exposure underneath the labels

Suppose a fictional $100,000 portfolio has $40,000 in Fund A, $30,000 in Fund B, $20,000 directly in Company C and $10,000 cash. There are four account lines. That count tells us little about the economic overlap underneath them.

Assume Fund A holds 25% of its value in Company C, while Fund B holds 10% in C. The portfolio's indirect C amounts are $10,000 and $3,000. Add the direct $20,000 holding and the total assumed exposure to C is $33,000, or 33% of the account. Counting only the direct line would miss 13 percentage points.

This simple look-through calculation assumes ordinary fully invested fund holdings, a common valuation date and no derivatives, borrowing or additional layers. Actual exposure analysis needs more detail when those assumptions fail. A published fund holding weight from last quarter is not necessarily its current weight, and a percentage of notional exposure is not automatically comparable with a percentage of net assets.

FINRA's concentration-risk explanation identifies overlapping direct and fund holdings, correlated investments and difficulty selling as ways a portfolio can be more concentrated than its account labels suggest. The general mechanics are useful internationally; this lesson does not import US account rules or FINRA's personalised-action suggestions. Source: FINRA, Concentrate on Concentration Risk, 15 June 2022; checked 10 October 2026.

Enlarge diagram

Different exposure views can overlap with each other

Now assume Company C belongs to Sector S, Fund A has 60% in S and Fund B has 20% in S. Sector S exposure is $24,000 through A, $6,000 through B and $20,000 directly: $50,000, or 50% of the portfolio.

It would be wrong to add the 33% Company C exposure to the 50% Sector S exposure and announce 83% of separate risks. C is already included inside S. These are overlapping views of the same portfolio, like counting a city's population inside a country's population. The classifications can both be useful without being additive.

Other useful views might group positions by currency, borrowing sensitivity, revenue source or geography. Each view needs a stated classification rule. A company's listing venue, headquarters and customer revenue locations are not the same geographic measurement. Calling a portfolio “international” without specifying the definition can conceal which question was actually answered.

This is why exposure analysis starts with a map rather than a large list of percentages. The map distinguishes mutually exclusive buckets from overlapping attributes. It can also show unknowns. If a fund's current holdings are unavailable, describing that gap is more informative than assuming its label supplies a complete exposure breakdown.

Worked example 1: a weighted scenario loss

Return to the $100,000 account. At the start, A is 40%, B is 30%, C is 20% and cash is 10%. Assume a single period with no deposits, withdrawals, rebalancing, distributions outside the measured returns or other trades. All returns use the same currency and period.

In Scenario One, Fund A falls 20%, Fund B falls 10%, Company C falls 35%, and cash has zero nominal return. The portfolio return is the sum of each starting weight multiplied by its scenario return.

Position	Starting value	Assumed return	Dollar change	Contribution to portfolio return
Fund A	$40,000	−20%	−$8,000	−8 percentage points
Fund B	$30,000	−10%	−$3,000	−3 percentage points
Company C	$20,000	−35%	−$7,000	−7 percentage points
Cash	$10,000	0%	$0	0 percentage points
Total	$100,000	—	−$18,000	−18%

The ending values are $32,000, $27,000, $13,000 and $10,000, adding to $82,000. Computing both the weighted return and the ending-dollar total provides a useful cross-check. An unweighted average of the three risky-asset returns would answer a different question and would ignore cash.

Do not apply the underlying Company C shock again to the fund values after already assigning complete fund returns. The specified fund returns are assumed to include all their holdings. The look-through map explains shared exposure; the scenario table supplies total position returns. Combining both as separate losses would double-count.

The table is also not an estimate of a likely future loss. No probabilities have been assigned to those returns. It answers a conditional question: if these assumptions occur together, what happens to this starting portfolio? That is useful without pretending it is a forecast.

Change the assumptions, then recalculate

In Scenario Two, change A to −30%, B to −20% and C to −50%, leaving cash at zero. The dollar losses become $12,000, $6,000 and $10,000. The total is $28,000, or 28%, and ending value is $72,000.

The portfolio did not change. The conditional outcome changed because the shocks changed. Calling the first calculation “the portfolio's maximum loss” would therefore be unjustified. A stress exercise requires a rationale for the selected shocks, and its omissions remain important even after the arithmetic is correct.

A further scenario might introduce wider execution costs or unavailable cash rather than simply making every return more negative. Another might affect one exposure while leaving others unchanged. These are different questions about vulnerability. A large table of arbitrary numbers does not become informative just because it contains many rows.

Enlarge diagram

The weights change after the returns occur

After Scenario One, Company C is worth $13,000 out of $82,000, so its new direct portfolio weight is about 15.85%. Cash is $10,000 out of $82,000, or about 12.20%. Neither change requires a purchase or sale. Different position returns change the denominator and therefore change weights.

This matters when linking several periods. Starting weights for the second period are not automatically the original 40/30/20/10 weights. Reusing the original weights would implicitly assume a rebalance, and a rebalance is a transaction assumption that should be stated. Cash flows create additional complications for measuring performance rather than risk.

The distinction is easy to test with dollar values. If the first period ends at $82,000, the second period begins with those actual position values unless the model records a trade or external flow. A spreadsheet that resets weights without recording the adjustment can accidentally teach a strategy different from the one described.

For a simple hypothetical 18% loss, returning from $82,000 to $100,000 requires about a 21.95% gain on the reduced base. The loss and recovery percentages use different denominators. This arithmetic does not predict recovery or how long it would take.

Worked example 2: correlation changes a volatility estimate

Volatility is a statistical measure of variation in returns. It is not the same as a scenario loss. For a two-asset model with finite variances, portfolio variance combines each weighted asset variance with a covariance term:

Portfolio variance = wA² × σA² + wB² × σB² + 2 × wA × wB × σA × σB × ρ.

Here w denotes weight, σ denotes return standard deviation, and ρ denotes correlation. All inputs must refer to compatible return definitions and horizons. Portfolio volatility is the square root of portfolio variance. The formula is a variance identity under its stated model, not a requirement that returns follow a normal distribution.

Assume a separate portfolio with 50% in each of two assets. Let their same-horizon volatilities be 20% and 10%. At zero correlation, the variance is 0.25 × 0.04 + 0.25 × 0.01 = 0.0125. Its square root is approximately 11.18%.

Now hold the weights and individual volatilities fixed but change correlation to 0.8. The covariance contribution is 2 × 0.5 × 0.5 × 0.2 × 0.1 × 0.8 = 0.008. Total variance becomes 0.0205, with volatility approximately 14.32%. The holdings have not multiplied; the assumed relationship between their returns changed.

This calculation demonstrates sensitivity to one input. It does not say every market crisis makes every correlation 0.8, or that the next loss will equal 14.32%. In an actual stress period, individual volatilities and exposures can change too. Holding them constant here isolates the covariance effect rather than describing an entire real event.

Correlation has precise limits

Zero correlation does not generally mean independence. It means the measured linear relationship is zero in the specified model or sample. A nonlinear relationship can remain, and both assets can lose in the same period. Likewise, perfect positive correlation means a perfect positive linear relationship, not necessarily identical percentage changes when volatilities differ.

Consider the abstract equally likely values X = −1, 0 and 1, with Y = X². Y is completely determined by X, but the covariance is zero: the negative and positive X contributions cancel. With nonzero variances, correlation is therefore zero. This is a mathematical counterexample to equating zero correlation with independence, not a model of investment returns.

Historical correlation also depends on choices such as sampling frequency, date range and how missing prices are handled. An estimate from monthly observations need not match one from daily observations. Neither is automatically fraudulent or wrong; they measure different datasets. A risk report should identify those choices before presenting the estimate as meaningful.

Investor.gov explains diversification as spreading exposure across and within investments, while noting that narrow funds may not provide broad diversification. That principle supports inspecting the holdings rather than relying on the number of funds. It does not promise a particular realised correlation or protection from every loss. Source: Investor.gov, Asset Allocation and Diversification; verified 10 October 2026.

Liquidity and a deadline can change the question

Suppose the first model portfolio is worth $82,000 after Scenario One, but a hypothetical payment of $15,000 is due before another asset sale would normally settle. Only the $10,000 cash line is assumed immediately usable. The portfolio is worth more than the payment, yet there is a $5,000 timing gap.

Resolving that gap requires facts about saleability, execution price, settlement and account access. It cannot be resolved simply by pointing at the $82,000 total. A quoted or estimated price is not a guarantee that the required quantity can be converted to available cash by the deadline.

For a second conditional illustration, assume a position marked at $5,000 can be sold in time, but only for $4,000 net in the stated stress scenario. Combined with $10,000 cash, that produces $14,000, still $1,000 short. The 20% discount is an invented assumption, not a typical haircut for any named product.

These examples separate market-value risk from cash-access risk. They also show why a liquidity assumption should be explicit in a scenario. Cash itself has inflation, currency and provider-related considerations; the zero-return cash line here is a modelling convenience, not a claim that every cash arrangement is risk-free.

Leverage and currency require a consistent denominator

A separate fictional account has $20,000 equity and borrows $10,000, supporting $30,000 of assets. If assets fall 10%, their value becomes $27,000. With debt still $10,000, equity becomes $17,000. The $3,000 loss is 15% of the original equity, before interest, fees or liquidation effects.

Applying the asset return directly to account equity would miss the borrowing. The example does not model margin terms, options or short positions; those can introduce additional nonlinear exposures and obligations. It merely shows why an asset-weight table that ignores liabilities is incomplete for a leveraged account.

Currency is another potential mismatch. If an investment rises 10% in dollars but the dollar's value falls 15% against the reporting currency, the combined unhedged return in that currency is 1.10 × 0.85 − 1 = −6.5%, before costs. Adding 10% and −15% gives a rough approximation, not the exact compounded result.

This applies whenever the account's reporting currency differs from the investment's currency; it is not specific to a UK reader. Hedging, underlying business exposures and fund structure can complicate the calculation. The example states an unhedged conversion assumption so the result has a clear meaning.

Attribution is different from a recommended change

In Scenario One, Fund A contributed eight percentage points to the 18% portfolio loss, while Company C contributed seven. Fund A's percentage return was less negative than C's, but its starting position was twice as large. Ranking positions by their individual percentage losses would therefore give a different ordering from ranking their dollar contributions.

This attribution does not establish what should be sold, bought or held. It describes how one set of assumed returns combines with one set of starting weights. A different scenario could produce a different ranking. Turning the largest contributor in one selected scenario into an automatic trading instruction would add a decision rule the calculation never justified.

Attribution also depends on the chosen level of analysis. The position-level table treats each fund as one line. A look-through decomposition could examine the companies inside those funds, but then the fund-level contributions must be replaced by their components rather than added to them. The total must still reconcile to the same portfolio change.

To see the difference between exposure and loss contribution, imagine a separate scenario where Company C has a zero return and every other underlying position changes. Its starting exposure could still be 33%, yet its direct contribution from that specific price move would be zero. Exposure describes how much of something is present; contribution describes the effect under a particular realised or assumed movement.

Document a scenario so somebody else can reproduce it

A useful scenario record contains enough information for another reader to reproduce the result without guessing. For the first worked example, that means the four starting dollar amounts, their common valuation date, the four returns, the single-period horizon and the absence of trades or outside flows. “Equities down” would not be precise enough to recover the $18,000 calculation.

It also distinguishes fixed assumptions from quantities derived by the calculation. The 40% starting weight in Fund A is an input. Its $32,000 ending value and its resulting share of the $82,000 total are outputs. If a spreadsheet overwrites an output and continues calculating, the result may no longer describe the original scenario.

An uncertainty note can be just as concrete. The current underlying fund weights might be unavailable, an asset might not have a reliable sale price, or the deadline might precede settlement. These are specific information gaps with specific consequences. They are more useful than a generic statement that “anything can happen.”

Finally, preserve the date and the version of the assumptions. Reusing a scenario six months later with different holdings can be a valid new exercise, but it is not the same test. A changed result could come from new positions, revised shocks or a different reporting currency. Recording those changes allows the reader to explain the difference rather than mistaking it for a calculation error.

What a useful risk conclusion contains

A reproducible risk explanation identifies the positions and liabilities, maps overlapping exposures, states the valuation date and currency, and separates the chosen measures. It then explains which assumptions drive a scenario or statistical result and what was left out. Precision in the output cannot compensate for an undefined input.

The examples here show three different outputs: 33% exposure to one fictional company, an 18% loss under one chosen scenario, and 11.18% volatility under a separate two-asset model. These are not interchangeable risk scores. They answer different questions about different assumed portfolios.

Understanding those distinctions is the learning objective. No calculation selects an allocation for the reader, guarantees diversification or defines a complete worst case. Next, measuring portfolio performance examines how to describe realised results; a historical result and a forward-looking risk assumption remain different kinds of evidence.