Graveyard / Positioning / US stocks
COT positioning on stocks and bonds
Does the CFTC Commitments of Traders report predict S&P 500 or Treasury returns?
The idea
The CFTC publishes how futures traders are positioned every week. Large speculators are said to be wrong at extremes. When their net position is low compared with the last three years, go long.
How we tested it
| Data | Legacy and traders-in-financial-futures reports. Equities: S&P 500 non-commercial net position as a z-score over 156 reports, long SPY while it is below zero. Bonds: T-bond and 10-year note speculator z-score, long TLT or IEF. Decisions at Monday's close after the Friday release. |
|---|---|
| Period | SPY: 1993 to 2026 (leveraged-fund and asset-manager variants 2009 to 2026). Bonds: October 2002 to October 2026, split at December 2021. |
| Costs | 1 bp per side, 6.5 bp per side as a stress case. |
| Holdout | Full, in-sample (80%) and out-of-sample (20%) Sharpe. |
| Control | Buy-and-hold on the same ETF is the comparison. SPY buy-and-hold Sharpe: 0.65 (1993-2026), 0.91 (2009-2026). |
Result
| SPY, speculator z below 0 | 84 trades, Sharpe 0.45, out-of-sample 0.44 (18 trades) |
|---|---|
| SPY, leveraged funds (2009-26) | 73 trades, Sharpe 0.65 |
| SPY, asset managers, follow | 40 trades, Sharpe 0.86, out-of-sample 1.49 (3 trades) |
| T-bond to TLT | 65 trades, Sharpe 0.30, out-of-sample −0.46 (10 trades) |
| 10-year note to IEF | 57 trades, Sharpe 0.47, out-of-sample −0.34 (13 trades) |
| TLT, 2002-14 vs 2015 on | Sharpe 0.60 vs 0.05 |
Why it failed
On stocks the signal adds nothing to buy-and-hold: the best-looking variant is mostly long SPY in a bull market, below the 0.91 buy-and-hold figure, and its out-of-sample period has three trades. On Treasuries it worked during the 2002-2014 bond bull market only. Since about 2015 the basis trade dominates leveraged-fund Treasury shorts, so their positioning no longer measures sentiment.
What would make it worth retesting
Not as a level signal. Positioning from a second source, or a signal built on changes in positioning, would be a new test.
Test id: cotspec-spy-event-v1, cottsy-tlt-event-v1. Past results of a backtest say nothing certain about the future.