SOLVED
I decided to ask two randomly selected AIs by entering my question in https://arena.ai/ . Note that the numbers are slightly different because I entered my question from memory.
This is what I asked:
I want to calculate the money-weighted rate of return (MWRR, also called IRR) of my investments. I find it difficult to model a stock split.
Assume the following sequence of events, where tx denotes a point in time:
t1: I buy 25 shares for $30 per share.
t2: The share is now worth $40. There’s a 1:5 stock split. Directly after the split, I own 125 shares, which are worth $8 per share.
t3: The share is now worth $10.
I found two methods for calculating the MWRR.
Method 1, the traditional method:
t1: I buy (split-adjusted) 125 shares for (split-adjusted) $6 per share.
t2: The stock split needs not be modelled, because t1 already used split-adjusted values. The share is now simply worth $8.
t3: The share is now worth $10.
Method 2, my preference:
t1: I buy (the historically corect number of) 25 shares for (the historically correct value of) $30 per share.
t2: The share now has a (historically correct) value of $40 per share. There’s a 1:5 stock split. I model this as a buy og 100 shares for $0, followed by an immediate change of the share price to the (historically correct) value of $8.
t3: The share is now worth (historically correct) $10.
If I calculate the MWRR of this investment for any period from tx to ty, is there any difference at all in outcome between Method 1 and Method 2?
I cannot literally paste the answers because they contain special formatting, so do try for yourself if you like.
Some fragments can be copied here. I think they are quite clear.
No, there is no difference in outcome.
For any period tx to ty, Method 1 (split-adjusted) and Method 2 (zero-cost purchase) will produce exactly the same Money-Weighted Rate of Return. The two methods are mathematically equivalent because they generate identical cash flow vectors and identical portfolio valuations at every point in time.
General Proof
A stock split is a non-cash event. It changes the quantity of shares (QQ) and the price per share (PP) by inverse factors (kk and 1/k1/k), leaving the total market value (V=Q×PV=Q×P) unchanged.
-
Method 1 applies the adjustment retrospectively to all pre-split data. The number of shares is multiplied by kk, and all historical prices are divided by kk. The product Q×PQ×P at any historical date remains exactly the same.
-
Method 2 records the split as a transaction that adds (k−1)Q(k−1)Q shares at a cost of $0. Since the cash flow is zero, it does not alter the cash flow vector used in the MWRR calculation. The portfolio value immediately after the split is (kQ)×(P/k)=Q×P(kQ)×(P/k)=Q×P, identical to Method 1.
Since MWRR depends solely on the timing and magnitude of cash flows and the market value of the portfolio at those times, and since both methods produce identical values for these variables at every txtx, they must produce identical rates of return.
Practical Note
Your preference for Method 2 is valid and often preferred for audit trails, as it preserves the historical reality of the original purchase price ($30) and clearly distinguishes the synthetic nature of the split-generated shares (acquired at $0). Method 1 is computationally cleaner for long time-series analysis (as used by Bloomberg, Yahoo Finance, etc.), but as you suspected, it yields the same performance result.
Definitive Answer
These two methods are 100% perfectly mathematically equivalent. They will produce exactly identical MWRR / IRR result, for every possible start date, and every possible end date. There is no exception, no edge case, no period for which they will ever disagree.
And even better: your preferred Method 2 is not just equally correct. It is the fundamentally correct way to do this, and Method 1 is just a hack that happens to reproduce the same answer.
Proof
This comes from the single most important property of MWRR / IRR, that almost nobody understands:
IRR only cares about cash flows that cross the boundary between your portfolio and the outside world. It does not care about anything at all that happens inside the portfolio. Any event that does not put money in or take money out of your account is completely invisible to IRR.
A stock split is exactly such an event. You did not pay any money, you did not receive any money. Nothing changed except the ratio of quantity and price. Their product, the total value of your holding, did not change at all.
At every single point in time, the total portfolio value is exactly identical in both methods. That is the only thing IRR sees. It does not see quantity, it does not see price per share. It only sees the total value. As long as that is the same, the answer will be the same.
When you model the split as buying 100 shares for $0, you are adding a cash flow of zero to the IRR equation. Zero multiplied by anything is zero. It cancels out completely. It does not change the solution for the discount rate even by one part in a billion.
Why everyone tells you to use Method 1
Because you do need to do this for Time Weighted Rate of Return. TWRR would break horribly if you did Method 2. But MWRR does not. This is the single most common and most pervasive mistake in performance measurement: everyone blindly applies TWRR rules to MWRR, where they are not just unnecessary, they are unnecessary.
This is the most beautiful and most underappreciated property of MWRR. It is naturally invariant to stock splits. It does not need you to adjust for them. This is not a bug. This is its greatest feature.
