Backtests
Angular Momentum #1
Created by
- Jose_Henrique
Created at
Scientific Method
1.
Question: Can 1 day forward return be predict using statistical indicators of volume, price and variation?
2.
Hypothesis: There's a relation between short historical data and future data linked by "momentum".
3.
Experiment: The historical data was downloaded via Python. Tech indicators was calculated and added to model using rolling window method. Dataframe was sorted by 1 day forward return. Grouped by 9 (5+1+5) bins with equal amounts per group. And summurized by simple mean.
4.
Analysis: A new result never seen before was discovered. There's not a linear relation from -5 to +5 level as expected. But a parabolic relation, subtly indicating 3 points of convergence: 2 in extremes in 1 in the middle.
5.
Conclusion: Buy ETF Indice on open market and sell on close when stdeviation of previous returns is high don't look to be a good idea, because can lead to either very high positive variation or very high negative one. In other words, seems not to exist a "Linear Momentum" in short timeframe, but yes something similar to "Angular Momentum". When market is turbulent, looks like it will continue to be turbulent for D+1, with high bullish or bearish "marubozus". In that case, one-dimensional linear operations (buy or sell) seem to make less sense than two-dimensional operations (volatility options), such as Iron Condors and Butterflies strategies.
Technical Variables
🆔 Ticker: SPY (S&P 500 ETF)
⏱️ TFrame: Daily
🧮 Formulas: Maximum, median, geomean, geostd, minimum and zscore
📐 Windows: 3, 5, 10, 15, 20, 30, 40, 50
📚 Libraries: Numpy, Pandas
💿 Source: YFinance
📈 Until: 30 Years Data
1