Linked returns and external cash flows

What you will learn

Derive cash-flow-neutral geometric returns and a unitization certificate, then construct a contribution-timing counterexample separating linked performance from chip gain.

Prerequisites: Net payoff and sunk costs · Variance and standard deviation · Inventory, cash and marked wealth

Declare the portfolio boundary and the clock

A larger ending balance can come from investment gains, new contributions, or both. Before dividing anything, define which chips belong inside the measured portfolio. A contribution crosses that boundary from the owner into the portfolio; a withdrawal crosses in the other direction. Buying an asset using cash already inside is an internal transfer. A dividend retained inside is investment income, not an owner contribution. The same physical transfer can receive a different classification if the measurement boundary changes, so a report must state that boundary explicitly.

Assume a fictional fully invested chip account begins with 200 chips and experiences two consecutive, equal-length periods. Its stipulated gross factors are 5/4 and 4/5, meaning returns of positive 25 percent and negative 20 percent. There are no fees, taxes, debt, price impact or distributions outside the account. Contributions buy exposure immediately; withdrawals sell exposure at the same fair value. Fractional bookkeeping units are permitted. Every flow has a known instant and a reliable valuation immediately before and after it. Rates are stipulated counterfactuals unaffected by account size. Nothing here estimates future returns or recommends an investment.

Build subperiod factors from values that bracket each flow

Let B1 be the value entering period one, E1 its value immediately before the boundary flow C, and B2=E1+C the value immediately after that flow. Positive C means a contribution; negative C means a withdrawal. Let E2 be the final value. The first factor is g1=E1/B1 and the second is g2=E2/B2. The linked time-weighted return is R=g1*g2−1. The second denominator includes the contribution because those chips are actually exposed in period two. The first numerator excludes it because adding owner money is not an investment gain.

For C=100 at the boundary, the ledger reads B1=200, E1=250, B2=350 and E2=280. Recover factors 250/200=5/4 and 280/350=4/5. Their product is one, so R=0. The account increased by 80 chips but the owner supplied 100; investment gain is 280−200−100=−20 chips. A zero linked return and a negative gain can coexist because different capital amounts experienced the two periods. These are different summaries of one valid ledger, not an arithmetic inconsistency.

Without a flow, adjacent numerators and denominators cancel: (E1/B1)*(E2/E1)=E2/B1. With a boundary flow they do not, because B2 is not E1. That is precisely why the ending-value ratio alone is insufficient. If only initial value, final value and total contribution are known, the intermediate valuation required for exact linking is missing. A chosen approximation may supply a model, but it cannot manufacture an observed valuation.

Derive a second certificate using units rather than balances

Give the initial account 200 bookkeeping units worth one chip each. After period one each unit is worth 5/4 chips, so the 200 units are worth 250. Issue C/(5/4)=80 new units for the 100-chip contribution. There are now 280 units and the price stays 5/4: 350/280=5/4. If the contribution raised the unit price, existing owners would appear to earn a gain merely because someone added capital. Issuing units at the current price keeps that event neutral.

Period two multiplies the unit price by 4/5, bringing it back to one chip. The 280 units are worth 280. The original 200 units finish worth 200, while the 80 new units finish worth 80 against their purchase cost of 100. Summing these two cohorts gives investment gain 0−20=−20. This independently verifies the balance calculation. The unit-price ratio from beginning to end is one, independently verifying the linked return. Units are a measurement device here, not securities being offered or a claim about a real fund.

A withdrawal is the reverse operation: redeem units at the current price. At boundary C=−100, redeem 80 units, leaving 120 units worth 150 before period two and 120 afterward. Add back the 100 withdrawn chips when reconciling the owner: 120+100−200=20 chips gained. The surviving unit price again ends at one. A positive withdrawal-adjusted gain is therefore compatible with the same zero linked performance.

Prove what changes when flows or period order move

For a general initial capital W, boundary contribution C and positive factors a and b, final value is V=(Wa+C)b=Wab+Cb. Subtract W and C to obtain gain G=W(ab−1)+C(b−1). This separates the original capital cohort from the new cohort. With W=200, a=5/4, b=4/5 and C=100, G=200*0+100*(−1/5)=−20. If the same contribution arrives only after period two, it earns no return: V=Wab+C=300 and G=0. Linked return remains ab−1=0 in either experiment.

Now reverse the factors while leaving the contribution at the boundary. First 200 becomes 160; add 100 to obtain 260; then multiply by 5/4 to finish at 325. Gain is 325−200−100=25. Geometric linking still gives (4/5)*(5/4)−1=0. Order commutes for the product of these two scalar factors, but not for the gain of new capital inserted between them. Algebraically, reversed gain minus original gain equals C(a−b)=100*(5/4−4/5)=45. The two ledgers independently differ by 325−280=45.

This invariance is conditional: contributions must not change the stipulated factors. Real cash held uninvested, liquidity limits, transaction charges or changed exposures could change the factor path. Nor does the example establish that reversing an actual historical strategy is feasible. It is a controlled accounting experiment that isolates the effect of capital timing. A money-weighted return is a separate rate calculation sensitive to dated owner cash flows; the chip gain here is not an internal rate of return.

Distinguish geometric linking from adding percentages

Adding 25 percent and negative 20 percent produces five percentage points, not the two-period return. The missing term is their product: (1+r1)(1+r2)−1=r1+r2+r1*r2. Here r1*r2=−1/20, canceling the apparent five-point gain. A 20 percent loss requires a 25 percent gain to recover because the recovery starts from a smaller base. The arithmetic mean of these period returns is 2.5 percent; it is neither the total linked return nor the constant per-period growth rate.

For two equal-length periods with positive factors, a constant equivalent per-period growth factor is sqrt(g1*g2). In this example it is one, so the equivalent per-period return is zero. Do not annualize without knowing how many such periods constitute a year. Do not average accounts or securities by geometrically linking them either: linking describes consecutive exposure through time. Combining simultaneous assets requires portfolio weights and belongs to a later lesson.

Construct a flow-neutral certificate in the workshop

Choose a signed 100-chip flow or no flow, place it between the periods or after both, and reverse the two factors. The displayed ledger updates immediately. Construct a return certificate by choosing the first and second denominator from the displayed alternatives. In the boundary-deposit case, deliberately use 250 for the second denominator and audit: the feedback identifies counted contribution as performance. Then select 350, audit again and compare the factor product with the unit-price ratio.

Freeze a valid certificate as a comparison. Reverse the periods or move the flow, build another valid certificate and inspect changes in linked return and chip gain. Predict the 45-chip order effect before revealing the audit. Use a boundary withdrawal to test sign conventions, and a zero flow to recover ordinary telescoping. Edits clear the current certificate feedback but preserve the explicitly frozen comparison. Reset clears everything. The activity uses local visit state and does not award account progress or claim a saved cross-device result.

Positive denominators are essential. Emptying the account completely at a boundary leaves no capital exposed during the next period, so its return cannot be recovered by dividing zero by zero. A later contribution requires a stated restart convention. Our controls deliberately keep every account positive; the shared arithmetic rejects a full depletion, nonfinite values and malformed inputs rather than silently displaying zero return. The official GIPS handbook discussion of time-weighted returns supports cash-flow valuations and geometric linking. This original unit certificate and counterfactual exercise do not constitute a GIPS compliance assessment.

Certify what the record identifies

A cash-flow total and an ending balance can determine chip gain while leaving linked performance unidentified. Keep initial 200, boundary contribution 100 and final 280 fixed, but call the unknown pre-flow valuation K. The linked factor is (K/200)*(280/(K+100)). With K=250 it equals one. With K=200 it equals 14/15, giving return negative 1/15. Both positive-capital ledgers satisfy the same observed totals and produce gain negative 20. They differ in intermediate performance, so the totals alone cannot choose between their linked returns.

This is a constructive identification argument: two complete compatible ledgers disagree on the quantity being requested. It is stronger than merely saying more data might be useful. Ask for the reliable pre-flow valuation and the exact flow clock. Neither a more precise final balance nor an extra decimal in the total contribution substitutes for that missing boundary observation. If an estimate must be used, label the valuation convention and test its sensitivity before comparing performance. The workshop supplies its intermediate values explicitly, which is why its certificate can be exact.

Hands-on: Build a flow-neutral return certificate

Move signed owner flows across the period boundary, construct exposed-capital denominators and freeze a paired unit and gain comparison.

Common mistakes

• Counting a contribution as earned return.
• Subtracting a contribution only at the end when it was exposed earlier.
• Adding period returns instead of multiplying gross factors.
• Calling chip gain a money-weighted percentage or IRR.
• Issuing units at the initial price after the account price has changed.
• Dividing by zero after full withdrawal or assuming missing valuations.

Keep these ideas

• Bracket flows with pre-flow and post-flow valuations, then link subperiod factors.
• Issue or redeem units at current value to keep an external flow price-neutral.
• G=W(ab−1)+C(b−1) isolates original and new capital cohorts.
• Identical linked performance can coexist with different chip gains; scope the metric before ranking outcomes.

Further reading

• GIPS Standards Handbook for Firms: Provision 2.A.24 discussion ↗

  Read Provision 2.A.24 and discussion (HTML lines 1625–1651), especially geometric linking, valuation at every external flow for a true TWR and valuation-error limitations. Also read the money-weighted IRR distinction (lines 1783–1801). Supports methodology only. Our chip ledger, unit proof, cohort identity and timing counterexample are original. No compliance assessment, source example or video reused. Checked 2026-10-25.