Research lab
Test ideas the way a quant would: a signal against random entries, a rule with costs, a tuning game that shows what overfitting costs, and the data to take into Python.
Simulated data from the course's market engine, not real prices. Results are in percent of price or in R; nothing here says a rule will work in real markets.
Trading real money can lose more than expected; nothing here is advice.
Base-rate lab
Pick a signal and a holding period. The lab compares what happened after the signal with what happened after every day in the same history, and with 1,000 random samples of entry days.
Running the numbers…
10 years of the no-edge market: it has no edge by construction, so any edge found here is luck.
Rule builder
Combine an entry signal, a filter and exits, then backtest on 10 years of the lab's trending market. Signals read candles up to their close and fill at the next open, and costs are always on. This is not the capstone's history, so nothing here can tune a rule on its locked years.
Simulated data, the same history for every learner. A good backtest here is a reason to test further, not a result.
Overfitting game
Tune a rule on the first 6 years of a new market as often as you like. When you are ready, reveal the last 4 years once. The score rewards a rule that holds up, and costs points for every block you switch on and every extra variant you try.
Round 1. Best score in this browser: -.
Costs are fixed at -0.05R a trade in this game, so a cheaper assumption cannot buy a better score.
Scoring: 100 points, plus 200 per 1R a trade in the hidden years (weighted by trades / (trades + 30), so a small sample counts for less), minus 100 per 1R the first years flattered the rule, minus 10 per block (trend filter, target), minus 5 per variant after the first. Never below 0.
Take the data to Python
Download the base-rate lab's 10-year history as CSV, with synthetic weekday dates, and a starter Jupyter notebook that runs a base rate against random entries, a bootstrap interval and a backtest with costs.
Put both files in one folder and open the notebook in Jupyter, VS Code or Google Colab (upload the CSV with it). It needs pandas and numpy, and makes no network calls.