Diversification Beyond the Number of Holdings

Two ETFs, two individual stocks, and cash hardly look like one basket. Yet bad news at one company sends both funds and one direct holding down together. Why do several account rows behave like the same judgment?

The rows show wrappers; risk comes from what is inside them. Open two fictional funds first, then inspect four-period return tables to see why historical offsets do not promise offsets under stress.

What you will check

Combine fund-mediated and direct company exposure, interpret correlation from consistently timed returns, and save a diversification review with joint stresses, unknown fields, and review conditions. The central task is identifying risk sources, not memorizing a statistical derivation.

Diversification aims to avoid concentrating money in one risk source. Investor.gov distinguishes diversification across and within asset classes and warns that narrowly focused funds may not diversify a portfolio.[1] More names begin the review rather than complete it.

Retain G47's funding boundaries and G74's CNY households unchanged. The following C3-49 dollar snapshot is separate: it does not reduce G47's candidate portfolio from $12,000 to $10,000 or replace its X holding. Continuous rebalancing belongs to G50's separate C3-50 ledger. Fictional fund components do not establish real ETF holdings or data permission.

Four investments contain five companies

At October 29, 2026, this synthetic long-only, unleveraged portfolio is worth $10,000. It holds Fund A $4,000, Fund B $3,000, direct X stock $1,000, direct K stock $1,000, and cash $1,000. The supplied fund components are complete, without other holdings, derivatives, or internal cash.

Look-through exposure multiplies the value of each fund holding by its internal company weight, then adds direct ownership. X contributes $1,600 through A and $1,500 through B, plus $1,000 directly: $4,100, or 41% of the portfolio. Adding fund percentages without their values, or considering only the direct 10%, would answer the wrong question.

Y contributes $1,200 plus $600, totaling $1,800. Z is $1,200, W $900, K $1,000, and cash $1,000. These six amounts still total $10,000. Looking through reclassifies assets; adding the funds again would count them twice.

Assume X and Y are technology companies and Z, W, and K are in other sectors. Technology exposure totals $5,900, or 59%. X's 41% is nested within that sector figure. They describe different levels and cannot be added to claim 100% risk.

Real fund research needs holding dates, stable asset identities, coverage, and weight definitions. A top-ten list supports only the known part. Unseen holdings are unknown, not proven independent. Funds of funds, derivatives, and leverage may require a different exposure analysis from this simple multiplication.

Shared risk extends beyond shared names

First stress X alone by −50%, leaving every other underlying asset and cash unchanged. There are no trades, costs, or taxes. The $4,100 X exposure loses $2,050, a 20.5% portfolio decline. Direct X loses only $500; the other $1,550 comes through funds.

Next start again from the original $10,000 for a separate joint stress: X and Y fall 40%, all other stocks 10%, and cash stays nominally unchanged. Technology loses $2,360 and the other $3,100 of stocks loses $310. Total loss is $2,670, leaving $7,330, down 26.7%. These stresses are alternatives, not consecutive periods.

Neither has an assigned probability or guarantees a maximum loss. They ask what happens if specified movements coincide. Different companies can still depend on interest rates, customer budgets, supply chains, financing, or liquidity. No duplicate name is required for a common driver.

FINRA identifies company, industry, geographic, employer, and liquidity concentrations, including exposures hidden inside funds.[2] Household income dependent on the same industry belongs in a joint household scenario. Wages are not another security value to insert into this portfolio's denominator.

Correlation describes co-movement, not labels

Here correlation means the sample Pearson correlation coefficient, measuring linear co-movement between two return columns. It ranges from −1 to +1. Positive values indicate similar directions of deviations from each column's mean; negative values indicate opposite directions. Zero linear correlation does not establish independence or exclude nonlinear relationships.[3]

Use a separate synthetic table of monthly strategy returns U and V. They are not measured results for ETFs A and B and are not inferred from the holding table. Currency, month-end intervals, and no-external-flow definitions match. No complete strategy validation is claimed.

Both means are zero. Deviations are exact opposites, giving sample correlation −1. Beginning every month half in each makes portfolio return their average, zero in all four months. This assumes restoring equal weights monthly and ignores costs. It demonstrates cancellation, not a free hedge available to buy.

Calculate correlation by subtracting each mean, summing row-by-row cross-products, then dividing by the square root of the product of both squared-deviation sums. Using percentage numbers 2 and −2, the numerator is −16 and each squared sum is 16, giving −1. Decimal inputs of 0.02 give the same ratio.

Both columns must vary for this denominator to be nonzero. The constant-zero portfolio has zero sample standard deviation. Its correlation with U is undefined, not zero. Correlation also does not directly report the amount earned or lost and is not causal evidence.

A different window can remove the offset

Use another separate four-month stress window. U returns −10%, −20%, −5%, and −15%; V returns −8%, −16%, −4%, and −12%. These are teaching inputs for a changed relationship, not a forecast derived from the previous table.

V is 0.8 times U in every row. Deviations therefore move proportionately together, giving +1 correlation. With the same monthly starting half weights, returns are −9%, −18%, −4.5%, and −13.5%. Every month loses. The previous window's negative correlation cannot override this one.

Two sample windows do not identify when actual market relationships change. The Cleveland Fed authors' working-paper page frames high crisis correlation and extreme losses as a research problem.[4] That is not a rule that every asset moves together in every crisis.

In real research, fix frequency, currency, overlapping dates, and missing treatment, then inspect multiple windows and stresses. Monthly correlation is not an intraday relationship. A long average can hide temporary co-declines, and four observations cannot guarantee estimation precision.

Check stale values too. Infrequent marks or suspended trading can flatten a return column. Zero-filling missing data can manufacture apparent low correlation. Examine quotation and valuation mechanisms before interpreting the statistic.

Three strategy names can be three routes to one exposure

Calling strategies trend, selection, and arbitrage does not separate their risks. They may hold the same stocks, depend on common funding and liquidity, or reduce positions together when volatility rises. Review assets, strategy returns, and operating channels separately.

Low historical return correlation cannot remove common data errors, a shared broker outage, or program failure. Full strategy-combination research requires verifiable individual records, account integration, and shared budgets, developed in G91/G92. This chapter identifies common sources without inventing new strategy capital.

Diversification also has costs: transactions, spreads, tax records, and ongoing review. Ten opaque funds may be harder to manage than fewer understood exposures. Calculate actual added work and trades, then identify which concentration was reduced. Holding count alone does not measure either benefit or cost.

Replace a fund and recalculate

Explanation question: How can ETFs from different issuers still produce 41% exposure to X?

Application question: An analyst treats everything outside a fund's disclosed top-ten holdings as independent assets. What is wrong? Write answers before checking the guide.

Different fund issuers do not change common underlying holdings. Multiply invested values by internal weights and combine them with direct shares. Undisclosed holdings are unknown, not verified nonoverlap. Obtain fuller evidence or limit the conclusion instead of letting gaps create diversification.

Change Fund B's components

Compare another static candidate snapshot. Replace all $3,000 of B with Fund C, whose complete holdings are X 0%, Y 20%, and W 80%. Keep everything else unchanged. Do not model the actual switch or its costs here. Recalculate X, technology, and the original joint stress. Has technology risk disappeared?

X becomes $1,600 through A plus $1,000 direct, or $2,600 and 26%. Y remains $1,800; technology totals $4,400 and 44%. Z is $1,200, W $2,400, and K $1,000, making $4,600 of other stocks.

The original joint shock loses 40% of $4,400 plus 10% of $4,600: $2,220, or 22.2%. This reduces specified company and sector exposure and that particular stress loss. It eliminates neither technology risk nor every other risk, and it does not establish improved net returns. Examine W concentration, common drivers, and switching costs. Renaming B without changing its holdings would change none of the original numbers.

Save a review of the common sources

Record date, currency, and market-value denominator. List direct and fund-mediated exposure with source, holding date, coverage, company amounts, and sector amounts. Add business, financing, liquidity, and household-income dependencies. Unknown fields need a route to evidence, not AI-invented constituents.

Retain return inputs, correlation definition, sample window, missing and zero-variation policies, alternative windows, and joint-stress assumptions. State accept, return, or awaiting evidence under G47's policy and G74's budget. Reducing concentration in this example is not funding approval.

Check that underlying amounts reconcile to original value without adding wrappers again. Hand-calculate one company and stress, and ensure low correlation has not become no-loss language. Revisit when holdings, income, or data quality changes, not merely when the displayed row count changes.

Count common risks before counting positions

Look through wrappers to companies, sectors, and business dependencies, then examine consistently defined return co-movement. Different names and historically low correlation do not independently establish diversification; missing fields limit the claim.

Combine changing correlation and simultaneous losses with funding policy. Additional assets or strategies warrant investigation when they reduce identifiable risks at acceptable cost and effort. More holdings alone cannot pass the review.

References

1. Investor.gov: Asset Allocation and Diversification. Diversification within and across classes and underlying overlap.
2. FINRA: Concentrate on Concentration Risk. June 15, 2022; companies, sectors, employers, liquidity, and look-through review.
3. NIST Dataplot: Correlation. Pearson definition and calculation; no significance test is used here.
4. Giuzio and Paterlini: Undiversifying during Crises: Is It a Good Idea?. Cleveland Fed WP16-28, 2016; authors' problem framing only, not a reproduced model or Federal Reserve position.

Sources checked October 30, 2026. Components, returns, and stresses are separate synthetic exercises, not real ETF history. AI-assisted draft; independent AI full-text and bilingual review is complete. Human financial review and target-reader testing have not been conducted and are not scheduled for this production round.