RESEARCH WORKSPACE · V1.0
RANDOMNESS LAB
Compare held-out historical returns with a simple bootstrap baseline. This is a statistical experiment, not a forecasting system.
1. Load historical data
Use the same CSV format as the dashboard: date,close. Files stay in this browser.
Waiting for files.
2. Method
Training · first 70%
Earlier daily returns form the sampling pool.
Held out · final 30%
Actual later-period return is compared with simulated paths of equal length.
Each simulated path independently samples training-period daily returns with replacement. This ignores time order, serial dependence, volatility clustering, and changing regimes. It is a deliberately simple randomness baseline—not a realistic market model.
3. Results
Run the experiment to see results.
Gold · actual held-out return
—
Gold · simulated median
—
Gold · actual percentile
—
S&P 500 · actual held-out return
—
S&P 500 · simulated median
—
S&P 500 · actual percentile
—
Gold distribution
S&P 500 distribution
| Asset | Train returns | Test returns | Actual max drawdown | Simulated median drawdown |
|---|
4. Interpretation limits
- The percentile only locates the held-out result within this particular simulated distribution. It is not a probability of a future outcome.
- A high percentile does not prove predictive skill; a low percentile does not prove the opposite is predictable.
- This evaluates a simple independent-resampling baseline, not a forecasting model or trading strategy.
- Historical results do not establish future performance. Do not use this experiment for financial decisions.