Comparing ARIMA and LSTM for Next-Day Nifty Price Forecasting

Summary

This project compares ARIMA time-series modeling with an LSTM neural network for next-day Nifty index price forecasts. The author used five years of adjusted closing-price data, selected an ARIMA specification after stationarity checks and parameter searches, and tested an LSTM with price and technical-indicator inputs. The report describes evaluation using price errors and directional predictions, and says ARIMA produced lower RMSE while LSTM did better at predicting direction. However, directional accuracy remained below 50% for both models, limiting its practical value.

The study is a single-index project with constrained data sourcing and computing resources, so its findings may not generalize to other assets or periods. It flags the risk of look-ahead bias when aligning predictions with next-day prices and notes that its simple long-only backtest assumptions, including buying at the close, may not be feasible in live trading. The project does not fully develop entry, exit, or risk controls, and calls for further testing, feature selection, and strategy refinement before production use.

Key ideas

• The project compares ARIMA and LSTM models for forecasting next-day Nifty prices.
• The ARIMA model had lower reported RMSE, while LSTM performed better on directional prediction.
• Both models had directional accuracy below 50% in the reported evaluation.
• The analysis used adjusted closing prices and added technical indicators as LSTM features.
• Prediction alignment, data quality, limited computing resources, and trading assumptions constrain the findings.

Tags

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