A breakout rule (M04 · 4.2) fired 400 times; 160 trades hit the profit barrier (+2R) and 240 hit the stop (−1R). A meta-model, trained with purged CV, would have taken 180 of the 400 signals, of which 99 were winners. (a) Compute the precision, recall and F1 of the primary rule (positive class: winning trade) and of the meta-filtered rule. (b) Compute total R earned by each. (c) Give two reasons the meta-model's historical improvement may overstate its live value.
Coding problems. The module's four exercises (m13_scoring, m13_garch, m13_hamilton, m13_purged_cv; in the app: Exercises → m13_garch, and so on) are part of this problem set: implement them until their tests pass. From a terminal: cd exercises && .venv/bin/python -m pytest modules/M13-forecasting/tests.