Shanghai Silver ML Timing Model Outperforms Buy-and-Hold with 62.5% Accuracy
A machine learning timing model for Shanghai silver futures, using XGBoost, achieved 62.50% directional accuracy over a 20-day horizon, with a long-short strategy net value of 7.73 versus 4.68 for buy-and-hold. The 5-day model underperformed during a bull market. Institutional research indicates the medium-frequency strategy generates excess returns after costs, integrating macro, price-volume, and event factors. High-confidence stratification produced only 12 signals in 20 days, limiting practicality.
IllustrationEditorial responsibility
- No named human review is recorded for this page.
- Reports are grouped by semantic similarity and deterministic rules. Language models may assist titles, summaries, translation and cross-source analysis; the page reads the event directly, while its address stays stable when the title changes.
- Summary covers the current reports
Cross-source coverage
Reporting timeline
Shanghai Silver Machine Learning Timing Model Effective; 20-Day Strategy Outperforms Buy-and-Hold
A financial news article from China Finance Network (CFi.CN) reports that a machine learning timing model for Shanghai silver futures (沪银) has demonstrated effectiveness. Data shows the 20-day XGBoost timing model achieved a directional accuracy of 62.50% and an AUC of approximately 0.618. The long-short strategy's ending net value reached 7.73, significantly higher than the buy-and-hold net value of 4.68. The 5-day model had 60.19% accuracy but underperformed simple holding and dual moving average strategies during the long bull market cycle. Recent institutional research indicates the 20-day timing strategy still generates significant excess returns after deducting transaction costs, suggesting medium-frequency signals are better suited to current silver volatility characteristics. The research notes the model does not pursue high-frequency capture but uses a fusion of macro, price-volume, and event factors to effectively filter short-term noise, providing stable entry and exit signals during periods of strong trend continuity. High-confidence stratification improves per-trade win rates but produces too few signals (only 12 trades in 20 days), limiting practicality. The default threshold 20-day strategy has a profit-loss ratio of 2.33, and combined with position management can control drawdowns. The core value for traders is providing an implementable medium-frequency timing framework, not replacing trend-following strategies.
Machine Learning Timing Model for Shanghai Silver Futures Shows 20-Day Strategy Outperforms Buy-and-Hold
An article from cfi_futures (China Finance Network) reports that a machine learning timing model for Shanghai silver (沪银) futures has demonstrated effectiveness. Data shows the 20-day XGBoost timing model achieved a directional accuracy of 62.50% with an AUC of approximately 0.618. The long-short strategy's ending net value reached 7.73, significantly outperforming the buy-and-hold net value of 4.68. The 5-day model had 60.19% accuracy but underperformed simple holding and dual moving average strategies during a long bull cycle. Recent institutional research indicates the 20-day timing model still generates significant excess returns after deducting transaction costs, suggesting medium-frequency signals are better suited to current silver volatility characteristics. The research notes the model integrates macro, price-volume, and event factors to filter short-term noise and provide reliable entry and exit signals during strong trend periods. While high-confidence stratification improves per-trade win rates, it produces too few signals (only 12 trades over 20 days) for practical use. The default threshold strategy offers a profit-loss ratio of 2.33, and with position management can control drawdowns. The core value for traders is providing an actionable medium-frequency timing framework rather than replacing trend-following strategies.
Shanghai Silver Machine Learning Timing Model Shows 62.5% Accuracy Over 20-Day Horizon
A research report cited by CFi.CN indicates that a 20-day XGBoost timing model for Shanghai silver futures (沪银) achieved a directional accuracy of 62.50% and an AUC of approximately 0.618. The long-short strategy's ending net value reached 7.73, significantly outperforming the buy-and-hold net value of 4.68. A 5-day model showed 60.19% accuracy but underperformed simple holding and dual moving average strategies during the long bull cycle. The report states that the 20-day timing model still generates significant excess returns after deducting transaction costs, suggesting medium-frequency signals are better suited to current silver volatility characteristics. The model integrates macro, price-volume, and event factors to filter short-term noise and provide stable entry and exit signals during trend-continuation phases. High-confidence stratification improved per-trade win rates but produced too few signals (only 12 trades over 20 days), limiting practicality. The default threshold strategy achieved a profit-loss ratio of 2.33, and with position management can control drawdowns. The report emphasizes the model's value as a deployable medium-frequency timing framework rather than a replacement for trend-following strategies.
Read sourceShow 2 older updatesHide older updates
Shanghai Silver Machine Learning Timing Model Shows 20-Day Strategy Outperforms Buy-and-Hold
According to data cited by CFi.CN, the 20-day XGBoost timing model for Shanghai silver (沪银) achieved a directional accuracy of 62.50% and an AUC of approximately 0.618. The long-short strategy's ending net value reached 7.73, significantly outperforming the buy-and-hold net value of 4.68. The 5-day model had 60.19% accuracy but underperformed simple holding and dual moving average strategies during the long bull cycle. Recent institutional research reports indicate that the 20-day timing strategy still generates significant excess returns after deducting transaction costs, suggesting medium-frequency signals are better suited to current Shanghai silver volatility characteristics. The research notes the model does not pursue high-frequency capture but instead integrates macro, price-volume, and event factors to effectively filter short-term noise, providing stable entry and exit signals during periods of strong trend continuity. While high-confidence stratification improves per-trade win rates, it produces too few signals (only 12 trades over 20 days), limiting practicality. The default threshold strategy achieves a profit-loss ratio of 2.33, and with position management can control drawdowns. The core value for traders is providing an actionable medium-frequency timing framework rather than replacing trend-following strategies.
Shanghai Silver Machine Learning Timing Model Shows 62.5% Accuracy Over 20-Day Horizon
According to data cited by CFi.CN on September 23, 2026, a 20-day XGBoost timing model for Shanghai silver (沪银) futures achieved a directional accuracy of 62.50% and an AUC of approximately 0.618. The model's long-short strategy produced an ending net value of 7.73, significantly outperforming the buy-and-hold net value of 4.68. A 5-day version of the model had 60.19% accuracy but underperformed simple holding and dual moving average strategies due to the sample period being a long bull market. Recent institutional research reports indicate that the 20-day timing model still generates significant excess returns after deducting transaction fees, suggesting that medium-frequency signals are better suited to current Shanghai silver volatility characteristics. The research notes that the model does not pursue high-frequency captures but instead integrates macro, price-volume, and event factors to effectively filter short-term noise and provide reliable entry and exit signals during periods of strong trend continuity. While high-confidence stratification improves per-trade win rates, it produces too few signals (only 12 over 20 days), limiting practical utility. Under default thresholds, the 20-day strategy achieves a profit-loss ratio of 2.33, and combined with position management can control drawdowns. The core value for traders is providing an implementable medium-frequency timing framework rather than replacing trend-following strategies.
Read source