How it works
What we publish, how we test it, and why you can check our work without ever seeing the code.
How signals are generated
Every signal comes out of a fixed pipeline. Nothing in it lets a person choose, edit or delay a signal.
Code only
Signals come straight from the strategy code. There is no form, screen or back door for typing in a trade.
No manual override
Once the code has produced a signal, nobody at oxrun can change it, including us.
Published on schedule
Everyone gets the same signal at the same moment. Weekly strategies send every Sunday at 8:00 AM ET.
How backtests are run
A replays a strategy over past market data as if it were running in real time. A backtest is only worth trusting if it plays by the same rules as live trading.
- Same code as live. The backtest runs the exact code that produces live signals, not a separate model of it.
- Only what was known then. Each decision uses only data available at that moment, which avoids .
- Realistic timing. A signal is followed by a trade at the next close, never at a price that had already gone.
- Costs included. Every trade pays a modelled , so results aren’t flattered by free trades.
- Real funds only. Strategies hold ETFs that actually traded, long-only and without leverage.
How we split history
We fix the rules using one slice of history, then test them on slices they never saw.
How robustness is tested
The biggest risk in any backtest is : rules tuned until they fit the past by accident. Robustness tests ask whether a result could be luck.
Held-back data
Rules are frozen before they meet later data. If they still work on and forward-test data, a fluke is less likely.
Resampling
We reshuffle the strategy’s own history into thousands of alternative timelines () to see the range of outcomes it supports, not just the one that happened.
Different market conditions
We check returns in falling, flat and rising markets, and across every rolling three-year window, to see whether the edge depends on one kind of market.
Statistical confidence
The estimates the chance the edge is real after allowing for track-record length and fat tails, and how much history is enough.
The robustness score
One 0–100 number summarises these tests, so strategies can be compared at a glance. The exact weighting will be published here. Current scores are preliminary.
Why the code stays private
The rules are the product of real research, and a strategy that everyone can copy stops working. So we keep the code private and prove the process instead.
- Protects the edge. If a strategy’s exact rules are public, crowds trade the same way and the advantage fades.
- You don’t need the code to check the process. Signals come from locked code, and every published signal is logged with a signature and timestamp (a ), so history can’t be edited afterwards.
- Judge the results, not the promises. The track record, the tests above and the signal history are all open for you to inspect.
See it in action
Browse the strategies and open any one for its full track record.