Testing Logistic and Linear Regression Strategies on the Nifty 50

Summary

This project builds daily trading signals for the Nifty 50 using logistic and linear regression. It derives predictors from OHLC relationships, including whether the close is nearer the high or low and whether the current high or low exceeds the prior day’s level. Logistic regression estimates the probability that the next close will be higher; the described strategy enters above a probability threshold and exits the next session, with an opening-price loss rule. A multiple linear regression model uses related inputs to predict the next close.

The study trains on one calendar year and tests on the following year, repeating the process across 2018–2021 to span pre-pandemic, decline, and recovery conditions. It reports that both models beat the Nifty 50 benchmark in five of six scenarios, with logistic regression delivering higher returns across phases and similar hit ratios. These are the project’s reported findings, not independently validated results. The analysis omits transaction costs and slippage, assumes fills at exact open and close prices, uses the index rather than futures, and does not implement position sizing.

Key ideas

• The study uses three binary OHLC-derived predictors to estimate the likelihood of a higher next-session close.
• Logistic regression produces a probability-based entry signal, while linear regression predicts a price value using related inputs.
• The models are trained on one calendar year and evaluated on the following year across 2018–2021.
• The project reports benchmark outperformance in five of six scenarios and better returns from logistic regression across phases.
• Results omit transaction costs and slippage, assume exact opening and closing fills, and exclude position sizing.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.