Every when-to-enter idea, tested and exhausted

The search for profitable patterns based on time and relative value has reached a dead end. I tested five distinct approaches ranging from session-specific reversals to statistical arbitrage, and none have shown the robustness required to survive transaction costs and out-of-sample (OOS) testing. In my previous research, I exhausted the search for price-based “shapes” in the data. This time, I shifted the focus to the clock: “when” and “who” is trading, rather than just “what” is moving. Using localized time data (Axiory EET server time synchronized to New York 09:30 ET), I ran the following five experiments.

Summary of Research 140-144

Why these strategies failed

• Tokyo Session Mean Reversion: I hypothesized that by avoiding the high-volatility London and New York sessions, I could capture mean reversion (MR) in the quieter Tokyo hours. While the win rate hovered around 56%, the strategy failed to beat costs. Even in low-liquidity zones, sudden trends combined with the M15 spread are enough to kill MR strategies.
• Index Turn-of-Month (ToM): While a genuine anomaly exists for US indices (US500/100/30) during the first four trading days of the month, the performance is marginal. Even at its best, the strategy essentially captures the underlying stock market beta. It overlaps with existing index-long strategies, offering no independent value.
• Cointegration (Pairs Trading): This was my strongest hope for low-drawdown, market-neutral returns. By trading the spread between two correlated assets, I aimed for mean reversion. While the in-sample (IS) results looked promising, the alpha completely collapsed out-of-sample. The spread correlations are fleeting and do not persist after accounting for costs.
• WMR London Fix: I tested for momentum and reversal patterns around the 16:00 London Fix. Both approaches failed to clear transaction costs. In fact, these results are optimistic because they do not even account for the typical spread widening that occurs during the Fix itself.

The Path Forward

The “price and clock” dimension is now effectively exhausted. My findings confirm that for the current data (FX, metals, indices), there are no robust edges left to find using purely technical or time-based indicators. I have decided against integrating external data such as news sentiment or macroeconomic reports, as my core philosophy remains that the market price already encapsulates all available information. Consequently, I will not be pursuing “outside” data sources. Instead, I will focus on two remaining avenues:

1. Refining existing sleeves: I will make minor adjustments to the existing calendar sleeve to incorporate the US-only turn-of-month effect found in Research 141.
2. Risk Management: I will continue to explore risk-scaling (volatility targeting) rather than searching for new logic. The search for new technical edges is not over, but it is clear that the “low-hanging fruit” of session and relative value timing has been picked clean. My next research phase will continue to dig deeper into the remaining but narrow corners of technical price structures.

How this connects

This verification builds on earlier ones (what failed before and what I tried this time, comparisons between approaches).

• Hunting exhaustively for a second source of returns
• Chasing 3% a month means a 70% drawdown, and every smart…
• Filtering trends by how information arrives: also no

Code to reproduce

You can reproduce this with the following scripts (see repo).

• scripts/research/sessutil.py