Overnight drift is real. Costs make it untradeable
The overnight drift is the phenomenon where the majority of stock market returns occur between the previous day's close and the next day's open.
The overnight drift is the phenomenon where the majority of stock market returns occur between the previous day’s close and the next day’s open. It is a genuine market anomaly, but it is effectively un-tradeable once you account for execution costs. I investigated this by decomposing daily OHLC (Open, High, Low, Close) data for major indices to isolate overnight performance versus intraday performance. The findings were striking:
| Index | Overnight Return (Annual) | Intraday Return (Annual) |
|---|---|---|
| US500 | +6.8% | +0.4% |
| US100 | +10.1% | N/A |
| US30 | +4.7% | N/A |
| For the US500, the overnight strategy yielded a Sharpe ratio of 1.35, which measures risk-adjusted return (the higher the number, the better the return relative to the volatility). More importantly, the strategy showed almost no correlation to my existing models. With a core correlation of +0.02 and a Connors correlation of +0.20, it offered a theoretically perfect way to diversify. In other words, it behaves completely differently from the strategies I currently run. | ||
| However, the strategy collapses under the weight of transaction costs. Because this approach requires entering and exiting a position every single day, you are looking at 250 round-trip trades per year. | ||
| Round-trip Cost (Basis Points) | Annual Return | Sharpe Ratio |
| :— | :— | :— |
| 0 bp | +7.2% | 1.25 |
| 1 bp | +4.7% | N/A |
| 2 bp | +2.2% | 0.37 |
| 3 bp | -0.4% | N/A |
| In a realistic environment, trading index CFDs at the open and close (periods of notoriously wide spreads) usually incurs costs of 2 to 4 basis points per round trip. At these levels, the annual return drops to below 2% or turns negative entirely. | ||
| This confirms the academic consensus that the “overnight premium” is extremely difficult to capture after costs. While the anomaly is real and uncorrelated, the combination of daily execution costs and the inevitable slippage during the market open and close makes it unsuitable for a stable prop-firm portfolio. Unlike the Connors-style approach, which relies on holding positions for several days to minimize the impact of fees, this high-frequency requirement is a dead end. I am leaving my current v1.5.0 system as is and moving on. |
How this connects
This verification builds on earlier ones (what failed before and what I tried this time, comparisons between approaches).