Products
Walk-forward validation with bootstrap confidence intervals, designed to help teams evaluate signal quality with institutional rigour.
Backtests are structured evaluations of how Snowtrail signals would have performed historically, tested against the specific question each product is designed to answer.
Every signal Snowtrail produces is accompanied by a corresponding backtest. Rather than asking "did this predict price direction?", each backtest is anchored to the signal's actual purpose -- for example, whether a stress regime correctly identified periods of elevated volatility and tail risk.
This ensures that validation is meaningful and that performance claims are grounded in what the signal was actually built to do.
Backtest against the question the product answers. A stress index should be tested against volatility and tail risk, not price direction. This principle ensures every evaluation is honest and relevant.
{
"product": "gbsi_us",
"signal": "system_stress",
"method": "walk_forward",
"results": {
"vol_multiplier": 1.42,
"tail_lift": 2.1,
"hit_rate": 0.68,
"bootstrap_ci_95": [1.18, 1.67],
"p_value_adjusted": 0.003,
"n_periods": 156
},
"robustness": {
"fdr_significant": true,
"min_sample_met": true,
"autocorr_adjusted": true
}
}Illustrative output. Fields and values shown for demonstration purposes.
The GBSI-US backtest evaluates whether the system stress signal correctly identifies periods of elevated volatility and tail risk in US natural gas markets.
Results include institutional metrics like volatility multiplier and tail lift alongside standard measures, with 95% bootstrap confidence intervals and autocorrelation-adjusted p-values.
Signal quality claims are only as good as the validation behind them. Most fall short of institutional standards.
Common problems with signal validation:
Testing against the wrong question (e.g. price direction for a risk signal)
In-sample results presented as out-of-sample performance
No confidence intervals, no statistical significance testing
Cherry-picked metrics that hide when signals fail
Snowtrail Backtests are designed to give you the full picture -- including where signals work, where they don't, and how confident you should be in the results.
Snowtrail Backtests support evaluation and decision-making at every stage.
Assessing whether a signal delivers meaningful edge before integrating it into a workflow.
Understanding how signal performance varies across different market conditions and regimes.
Examining false stress rates, missed stress periods, and worst-case drawdowns to understand risk.
Providing quantitative evidence for investment committees and compliance reviews.
Snowtrail Backtests are designed to meet institutional standards of rigour and transparency.
All backtests use expanding-window walk-forward methodology. No in-sample results disguised as predictions.
Block bootstrap confidence intervals, autocorrelation-adjusted p-values, and false discovery rate controls.
Volatility multiplier, tail lift, and decision lens analytics -- not just hit rate. Metrics designed for how trading desks actually evaluate signals.
Every backtest includes failure analysis: false stress rates, missed stress periods, and worst-performing windows. No cherry-picking.
If you'd like to review our backtest methodology and results for specific products, we'd be happy to walk you through the framework and evidence.
Request Demo