Diagnosing Missing-Data Failures Between Backtests and Paper Trading
Summary
This discussion addresses a strategy that runs successfully in backtesting but fails during a simulated live trial. A suggested cause is a missing-data processing step that removes every row because one or more columns contain empty values. The proposed diagnostic is to inspect node outputs in sequence, starting with the missing-data node and then checking earlier nodes to locate where the data disappears.
The replies also describe a possible warm-up issue: some features need observations from earlier days before they can produce values. Until enough history is available, those features may be empty, causing rows to be removed; the strategy may then begin working after the required history accumulates. The suggested remedy is to increase the backward data window in the feature extraction settings. These are hypotheses based on an error report, not a confirmed diagnosis, since the strategy source and exact error details are absent. A reply also suggests trying a one-day backtest as a check.
Key ideas
- A successful backtest does not rule out missing-data problems during simulated live execution.
- A missing-data step may remove all rows if required columns contain empty values.
- Inspecting successive node outputs can help identify where the data disappears.
- Features that need earlier observations may remain empty until their warm-up history is available.
- Increasing the backward data window is suggested, but the discussion lacks source code to confirm the diagnosis.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.