In our trading strategy we identify US stocks whose prices rise sharply within a short time and may therefore move well away from their fundamental company value. Such price movements can create attractive short opportunities. Not all of these situations, however, lead to equally favorable trading results.
With Ridge 75 we have developed a quantitative selection model that scores potential trades using historical price and trading data. The aim is to systematically filter out less promising opportunities and thereby improve the average quality of the remaining trades.
The model complements the existing trading rules. The underlying strategy and the defined stop-loss and profit-taking rules remain unchanged.
Methodology: statistical modeling instead of subjective selection
Ridge 75 is based on ridge regression, an established statistical method that can be used in particular where the data base is limited and explanatory variables are correlated.
The model analyzes patterns in historical price and trading data and uses them to estimate the relative attractiveness of potential trading opportunities. On this basis, it uses a defined selection threshold to decide which candidates are considered for trading.
We deliberately chose a comparatively simple modeling approach. The regularization of ridge regression limits the complexity of the estimated relationships and can thereby help reduce overfitting to historical data. Compared with more complex model architectures, the statistical approach is also comparatively easy to examine and verify.
The model is retrained regularly with new historical data so that its parameter estimates can be updated. Which features enter the calculation, how they are weighted and where the specific selection threshold lies is part of our proprietary trading methodology.
Backtest: results with and without Ridge 75
We examined Ridge 75 using historical data from March 2022 to September 2026. We compared two variants of our strategy: trading all candidates that meet the basic trading rules, and trading the same set of candidates after additional selection by Ridge 75.
Position sizes and assumed trading costs were set identically in both variants.
| Metric | Without Ridge 75 | With Ridge 75 |
|---|---|---|
| Average result per trade | +6.0% | +7.1% |
| Maximum account decline (max. drawdown) | 31.8% | 19.9% |
Over the period examined, the average result per trade thus rose by 1.1 percentage points. At the same time, the maximum historical account decline fell by 11.9 percentage points.
The results suggest that the additional selection improved the composition of the candidates traded. Instead of trading every stock that is suitable in principle, the strategy concentrates on those opportunities the model rates as more promising.
Supplementary control tests support this finding: in the backtest, Ridge 75 performed better than a random selection of candidates of comparable size. When trade results were randomly shuffled, no corresponding selection advantage appeared, as expected.
These tests provide additional evidence that the observed results are linked to the selection logic. They do not, however, rule out chance effects or overfitting to the historical data used.
Implications for the risk profile
The results show two changes relevant to the strategy: a higher average result per trade and a lower maximum historical account decline.
Both metrics have to be viewed in the context of the overall trading strategy. An improved average trade performance does not automatically mean a higher risk-adjusted return under all market conditions. Likewise, a lower maximum drawdown in the backtest is no guarantee of a correspondingly smaller decline in live trading.
Ridge 75 changes neither the underlying trading idea nor the existing exit rules. Its purpose is to improve the selection of the stocks traded before a position is opened.
Limits and ongoing validation
The results so far are promising but have to be placed in the right methodological context.
- Historical simulation: All metrics shown come from a backtest. They are not the results of a live test carried out entirely outside the model development.
- Model selection and overfitting: The model was selected and tuned on the same historical data on which the results shown were obtained. There is therefore a risk that part of the observed advantage will not carry over to future market conditions.
- Different market phases: In 2022 and 2023 the difference between the variants was small. From 2024 to 2026 the observed selection advantage was more pronounced. These differences underline that the effectiveness of the model may depend on the market environment.
- Comparability of the metrics: The results are based on a separate test setup and are therefore not directly comparable with the figures in our strategy paper.
We monitor the model's performance continuously and update the parameter estimates regularly with new data. What remains decisive is whether the selection advantage observed in the backtest is confirmed under real trading conditions and in different market phases.
Conclusion
Ridge 75 adds a systematic quantitative selection layer to our trading strategy. The backtests so far show a higher average result per trade and a lower maximum historical account decline compared with the variant without additional model selection.
Whether and to what extent these advantages can be realized in live trading remains to be seen. Ridge 75 should therefore be understood as part of a continuously monitored quantitative trading process, not as a guarantee of future investment results.
Past results are not a reliable indicator of future developments. A loss of the invested capital remains possible. This article is for information purposes only and constitutes neither investment advice nor an invitation to buy or sell financial instruments. Our Terms & Conditions apply.