Graveyard / Calendar effects / US stocks
Treasury auction cycle trading
Do Treasury bond ETFs rally after a Treasury auction?
The idea
Dealers need a price concession to absorb scheduled supply, so Treasury prices fall into the auction and recover afterwards. It repeats every month and is a payment for liquidity, not information.
How we tested it
| Data | 616 auctions of 7- and 10-year notes and 346 auctions of 20- and 30-year bonds, including reopenings, from the US Treasury's data. Long IEF (7 and 10 year) or TLT (20 and 30 year), half a sleeve each, from the auction-day close for 5 trading days. |
|---|---|
| Period | July 2002 to October 2026. |
| Costs | 1 bp per side; stress case 6.5 bp. |
| Holdout | In-sample, out-of-sample from 2022, and a volatility filter. |
| Control | The unconditional 5-day mean return. |
Result
| Sharpe full / in-sample / out-of-sample | 0.41 / 0.66 / −0.31 |
|---|---|
| Profit factor / max drawdown | 1.32 / −26% |
| Sharpe at 6.5 bp | 0.09 |
| Time in the market | 66% |
| TLT, 5 days into the auction | −28.6 bp (unconditional +8.5 bp) |
| TLT, 5 days after the auction | +7.7 bp |
| IEF, 5 days after the auction | +14.0 bp (t 3.6) |
Why it failed
The strategy is mostly long-duration exposure: it had a Sharpe of −1.27 in a bond bear market and +1.89 in a bull market. The concession before the auction is real for TLT, but there is no recovery after it. The volatility filter did not help.
What would make it worth retesting
Closed. A fresh pre-registered test that went short TLT into the 20- and 30-year auctions (and long after) had a Sharpe of 0.59 / 0.75 / 0.12, and on untouched 1980-2000 data it was flat (t −0.55), so the lead is closed.
Past results of a backtest say nothing certain about the future.