Research · October 2026
Is short interest still an alpha signal? Eight years of FINRA data, tested
We sorted 2,967 US stocks into deciles by days to cover and by the change in short interest, every FINRA report from 2018 to 2026, on survivorship-free prices. The most-shorted stocks lagged, but not by enough to be statistically reliable.
The short answer
Short interest is one of the best-known signals in equities. Academic work from the 2000s and early 2010s found that heavily shorted stocks, and especially those with many days to cover, went on to underperform. We tested both signals on every FINRA short interest report from 2018-01 to 2026-09: 209 twice-monthly rebalances, delisted stocks included.
- The most-shorted stocks lagged. The top decile by days to cover trailed the average stock by 2.3% a year in the broad universe and 1.1% in liquid names.
- But the gap is not statistically reliable. The t-statistics are between -1.1 and -0.5, well inside the range that chance produces.
- The least-shorted stocks did worst of all. The bottom decile by days to cover trailed the average by 6.8% a year, so the classic long-short (buy the least shorted, sell the most shorted) lost 4.5% a year. Days to cover divides short interest by trading volume, so the lowest decile is crowded with the most heavily traded names, and that, not short sellers' skill, drives its returns.
- It is expensive to trade. At each twice-monthly rebalance, 34% to 42% of the most-shorted decile by days to cover is new, and 84% for the change in short interest, before any cost of borrowing those shares.
Our reading: on its own, in a simple sort, short interest has not been a dependable source of alpha since 2018. That fits the wider evidence that published anomalies weaken once they are widely known. It remains useful as a risk input, as an ingredient in multi-factor models, and for event setups such as squeezes, which this test does not address.
Growth of $1 in the long-short portfolios
Results
| Signal and universe | Stocks per period | Long-short, a year | t-stat | Most-shorted vs average, a year | t-stat | Most-shorted turnover per rebalance |
|---|---|---|---|---|---|---|
| Days to cover, broad | 2,967 | -4.5% | -0.89 | -2.3% | -0.82 | 34% |
| Days to cover, liquid | 938 | -3.5% | -0.76 | -1.1% | -0.48 | 42% |
| Change in short interest, broad | 2,799 | -0.5% | -0.18 | -2.1% | -1.07 | 84% |
| Change in short interest, liquid | 936 | -0.1% | -0.03 | -2.4% | -0.97 | 88% |
Annualised from 24.1 rebalances a year. Long-short is decile 1 (least shorted) minus decile 10 (most shorted).
Return of each decile relative to the average stock, a year
| Signal and universe | D1 | D2 | D3 | D4 | D5 | D6 | D7 | D8 | D9 | D10 |
|---|---|---|---|---|---|---|---|---|---|---|
| Days to cover, broad | -6.8% | +1.6% | +3.0% | +0.9% | +3.0% | +0.1% | +0.6% | +0.2% | -0.3% | -2.3% |
| Days to cover, liquid | -4.6% | +3.1% | +4.6% | +1.0% | -0.3% | +0.2% | -2.5% | +0.8% | -1.1% | -1.1% |
| Change in short interest, broad | -2.6% | -2.8% | +0.9% | +2.4% | +1.8% | +1.4% | +0.9% | +2.2% | +1.3% | -2.1% |
| Change in short interest, liquid | -2.5% | -3.1% | +0.4% | +2.3% | +0.5% | +1.3% | +3.5% | +1.7% | +0.8% | -2.4% |
D1 is the least shorted (lowest days to cover, or the largest fall in short interest), D10 the most.
Long-short return by calendar year
| Year | Days to cover, broad | Days to cover, liquid | Change in short interest, broad |
|---|---|---|---|
| 2018 | -5.9% | -11.5% | +11.8% |
| 2019 | +4.4% | +9.3% | +8.0% |
| 2020 | -22.5% | -13.5% | -16.0% |
| 2021 | +5.6% | -27.8% | +2.4% |
| 2022 | +9.6% | +21.9% | +7.4% |
| 2023 | -6.6% | -1.4% | +1.3% |
| 2024 | +0.2% | +0.8% | +1.2% |
| 2025 | -19.5% | -2.2% | -11.3% |
| 2026 | -7.4% | -4.5% | -7.8% |
2026 runs to 2026-09-10.
How we tested it
The method was fixed before we looked at the results.
- Data. FINRA consolidated short interest for exchange-listed securities, twice a month, and FinzData's survivorship-free daily prices with splits and cash dividends. 1,281,355 stock-periods in all.
- No look-ahead. FINRA publishes about seven business days after each settlement date, so each portfolio is formed at the close of the first trading day at least 10 calendar days after settlement, and held to the next formation.
- Universes. Broad: exchange-listed stocks (ETFs excluded) priced at $5 or more and trading at least 100,000 shares a day. Liquid: $10 or more and at least 1,000,000 shares a day. Delisted stocks stay in until their last trade.
- Portfolios. Equal-weighted deciles. Change in short interest uses stocks with at least 100,000 shares short at the prior report.
- Limits. Returns are gross of trading and borrowing costs. A stock that delists earns its return to the last close and nothing after, since no delisting return is available. We have no float data in this test, so short interest as a share of float, a variant some studies prefer, is not covered here.
Run it yourself
import finzdata as fz
c = fz.Client()
si = c.short_interest(["GME", "AAPL"], start="2018-01-01") # every FINRA report
px = fz.Ticker("GME").history(start="2018-01-01") # survivorship-free pricesBrowse the latest report by stock and sector at finzdata.com/short-interest. The last 12 months are free with an API key; the full history since December 2017 comes with a look-back bundle, alongside survivorship-free prices, 13F holdings, insider trades and point-in-time macro. See pricing.
Built 2026-10-11. Research, not investment advice. Past performance does not predict future returns.