Selling Fast and Buying Slow: Heuristics and Trading Performance of Institutional Investors

Akepanidtaworn · Di Mascio · Imas · Schmidt — Journal of Finance 78(2), 2023 (NBER WP 29076)

Thesis

Even market professionals fall back on cheap heuristics in one domain: selling. Portfolio managers show genuine skill when buying but systematically destroy value when selling — their sells underperform a coin-flip among stocks they already own. The asymmetry comes from attention, not from constraints or flow pressure.

Method

Inalytics Ltd data: 783 long-only institutional equity portfolios, Jan 2000 – Mar 2016 (average portfolio $573M; ~51k portfolio-months; 2.4M buys, 2.0M sells).
Each trade judged against a random counterfactual: what if they'd bought (or sold) a different stock they already held that same day? Factor-neutral matching on size, value, volatility, momentum, prior returns.
Value-added measured as post-trade abnormal return differences per dollar traded; linked to actual benchmark-adjusted portfolio returns.

Key findings

Buys: +100 bps/year per dollar purchased vs the random-buy counterfactual. Clear skill.
Sells: −80 bps/year vs a factor-neutral random-sell counterfactual. Pros would literally do better selling a randomly-chosen holding. For scale: mutual fund fee ratios are 20–50 bps — they burn 2–4× the fee in bad selling.
Underperformance is worst for fundamentals-driven managers with concentrated, high-active-share portfolios; momentum traders show the least sell-side drag.
Mechanism — neglected ideas: sells are disproportionately drawn from extreme positions (very large or very small weights, big past movers), i.e. whatever is salient rather than what analysis says to exit. When sells avoid the extremes, the underperformance disappears.
Both skill measures predict actual fund returns: buying skill adds, selling drag subtracts — this is not a statistical artifact.

Limitations

Inalytics clients opted into monitoring — selection toward sophisticated, tax-exempt institutions; retail extrapolation is a guess.
Long-only, no leverage/shorts: the sell decision is "which position to exit", not "go short" — the setting is deliberately narrow.
The random-sell counterfactual is harsh in one sense (no sell could be forced by mandates) — though the authors split on outflow pressure and results hold.
2000–2016 predates the passive/AI era; no evidence on how quant tooling has changed the asymmetry.

So what for Nick

Selling is a different skill from buying, and even people paid millions don't have it. If he prunes positions by "what's on my mind" — the biggest loser, the shiny winner — he's running the exact heuristic that costs pros 80 bps.
The fix the data implies: make selling rule-based and forced-random-ish — e.g. periodic review of every position (not the salient ones), or pre-committed exit rules written at buy time.
When reading finance content, discount "when to buy" advice relative to "when to exit" advice — the exit side is where even experts have no edge.