The familiar description “sell winners too early and hold losers too long” contains a judgment about timing. A sale-versus-hold ledger measures a pattern; demonstrating that the timing was poor requires a separate benchmark for what the investor should have done.
A fictional desk review
Maya reviews ten long stock positions at one decision time. Each has one known purchase price, and all ten can be sold. Five are above their purchase prices and five are below. She sells three winners and one loser, leaving two paper gains and four paper losses. “Paper” means unrealised, not imaginary: those positions have already changed in market value.
This deliberately simplified ledger counts each position once, ignores fees, taxes and dividends, and assumes no splits or partial sales. Execution prices equal the prices used to classify the positions. It is a teaching example, not observations from a brokerage study.
Here the terms count positions, not money amounts:
A pattern does not identify a motive
What a useful audit must preserve
For a real trade history, make the ledger reproducible before interpreting its difference:
- Opportunity set and time window. State which accounts, instruments and dates qualify. Counting only sale days answers a different question from observing every trading day. Repeated observations of one unsold position are dependent; a large row count is not an equally large set of independent decisions.
- Reference prices and classification. Specify average purchase cost, individual tax lots or another reference, and how fees, dividends, currencies and exact break-even positions are handled. A bookkeeping cost basis is observable; the investor's psychological reference point may differ.
- Splits and partial trades. Reconcile share counts and per-share costs after corporate actions. Decide whether selling part of a holding counts as a sale event, how its retained portion enters the ledger, and whether observations are weighted by positions, shares or value. These conventions can change the statistic.
- Reasons for trading. Rebalancing can require trimming appreciated holdings. Taxes can favour realising some losses or deferring gains, with effects depending on account, jurisdiction and period. Liquidity needs, trading costs, restrictions and changed forecasts can also affect exits. A historical December pattern is not a universal tax calendar or a trading rule.
If a denominator has no opportunities, its rate is undefined, not zero. These are design choices and competing explanations to test, rather than corrections that mechanically remove psychology. Causal inference needs evidence beyond a positive PGR-minus-PLR difference. For Maya's desk, a useful next comparison is between recorded decisions and the desk's pre-existing exit criteria, risk limits and forecasts. Automatically reversing every winner sale would create a new strategy whose performance still needs evaluation.
Check the denominator
A second fictional audit observes eight winning positions and two losing positions once at a common decision time. Four winners and both losers are sold in full. There are no break-even positions. Does selling more winners demonstrate a greater propensity to realise gains under PGR and PLR?
Sources
- Shefrin, H. and Statman, M. (1985). The Disposition to Sell Winners Too Early and Ride Losers Too Long: Theory and Evidence, Journal of Finance 40(3), 777–790. Abstract and pp. 777–779: original framing and proposed mechanisms.
- Odean, T. (1998). Are Investors Reluctant to Realize Their Losses?, Journal of Finance 53(5), 1775–1798. §I.B, §III.A–B and conclusion: alternatives, opportunity-based measurement and interpretation.
- Barberis, N. and Xiong, W. (2009). What Drives the Disposition Effect? An Analysis of a Long-Standing Preference-Based Explanation, Journal of Finance 64(2), 751–784, especially pp. 751–758: the preference-to-trading link depends on model implementation.