# Indicator Training - Train, Optimize and Combine Custom Technical Indicators **Last Updated**: 2026-03-17 **Version**: 1.0.0 ## How It Works ### Open the Indicator Training Module Navigate to the Indicator Training page from the sidebar menu. The interface presents four distinct training modes accessible via tabs: Recommendations, Evaluation, Optimization, and Combination. ### Select a Training Mode Choose your training mode based on your objective. Recommendations uses AI to suggest indicators suited to your asset and timeframe. Evaluation tests how well specific indicators predict returns. Optimization tunes indicator parameters for peak performance. Combination merges multiple indicators into composite alpha signals. ### Configure Asset and Timeframe Select the target symbol using the company search field, set the analysis date range, and choose the data timeframe (1min, 5min, 15min, 1h, daily). These parameters define the historical dataset used for training. ### Set Optimization Targets and Constraints Define your optimization objective: Information Coefficient (IC), IC Information Ratio (ICIR), Sharpe Ratio, or cumulative return. Set constraints such as maximum drawdown tolerance and minimum sample size to prevent overfitting. ### Select Indicators for Training Browse and select from the library of 138 technical indicators organized by category: momentum, volatility, volume, trend-following, and mean-reversion. In Combination mode, select up to 5 indicators and define the maximum combination depth. ### Execute the Training Process Click Execute to start the training pipeline. The backend runs the selected mode against historical data, applying cross-validation where applicable. Progress is displayed in real time via the status panel. ### Analyze and Export Results Review training results including IC scores, rank correlations, parameter sensitivity charts, and equity curves. Compare candidate indicators side by side and export winning configurations for use in your trading strategies. > Indicator Training provides a systematic framework to evaluate, optimize, and combine technical indicators using quantitative metrics rather than subjective judgment. ## Tips & Best Practices - Always run Evaluation mode first to establish baseline indicator performance before attempting Optimization; this prevents wasting compute on indicators that have no predictive signal in your target asset. - When optimizing, prefer ICIR (Information Coefficient Information Ratio) over raw IC as your target metric. ICIR accounts for consistency of predictive power across time, reducing the risk of overfitting to a single favorable period. - In Combination mode, pair complementary indicator types (e.g., a momentum indicator with a volatility indicator) rather than combining indicators of the same family, which often share correlated signals and add little diversification. - Use at least 2 years of daily data or 6 months of intraday data for training to ensure the model encounters different market regimes including both trending and ranging conditions. ## Frequently Asked Questions ### What is the difference between Evaluation and Optimization modes? Evaluation mode tests an indicator with its default parameters and reports how well it predicts future returns using metrics like IC and ICIR. Optimization mode searches across a range of parameter values (e.g., lookback period, smoothing factor) to find the configuration that maximizes your chosen objective metric. Start with Evaluation to identify promising indicators, then use Optimization to fine-tune the best candidates. ### How does the AI Recommendations mode work? Recommendations mode sends your asset characteristics, timeframe, and recent price behavior to the LLM engine, which analyzes the data context and suggests indicators likely to perform well. The AI considers factors like current volatility regime, asset class, and historical indicator effectiveness on similar instruments. Results include a ranked list with rationale for each recommendation. ### What does the Information Coefficient (IC) measure? IC measures the rank correlation between an indicator's predicted signal and the actual forward returns. An IC of 0.05 or above is generally considered meaningful in quantitative finance. Higher IC values indicate stronger predictive power, but consistency (measured by ICIR) matters as much as magnitude. ### How can I avoid overfitting during optimization? The platform applies walk-forward cross-validation by default, splitting data into in-sample training and out-of-sample testing windows. Additionally, set conservative constraints such as minimum sample size and maximum drawdown limits. Avoid optimizing on very short date ranges, and always compare in-sample versus out-of-sample performance in the results panel. ### Can I use trained indicators in other modules like the Arena or LLM Driving? Yes. Once you identify an optimized indicator or combination, you can reference it in the Arena workflow builder or LLM Driving strategy configuration. The trained parameters are saved to your account and appear in the indicator selector across all compatible modules. ## Important Notes > Indicator optimization is performed on historical data and optimized parameters may degrade on unseen data. Over-optimization (curve fitting) is a significant risk. Always validate results on out-of-sample periods and use walk-forward analysis before deploying in live strategies. --- Source: https://stratcraft.ai/help/indicator-training/