"It Looks All the Same to Me": Cross-index Training for Long-term Financial Series Prediction
Stanislav Selitskiy
TL;DR
The paper investigates whether an artificial neural network trained on one major stock index can forecast another index with comparable or better accuracy over a 30-day horizon, as a test of the Efficient Market Hypothesis. It compares a broad set of ANN architectures, including ready-to-use layers and custom-designed components, across NASDAQ, Dow Jones, NIKKEI, and DAX using data from 2005–2022 and evaluates cross-training versus in-market training with MAPE and RMSE, augmented by Wilcoxon tests. Results show cross-training often yields similar or improved accuracy for several cross-market pairs, with LSTM sequence-to-vector being particularly robust while RBF networks exhibit higher volatility, leading to a nuanced, index-dependent view that provides weak support for EMH. The study highlights shared information across global markets accessible via ML, while noting limitations and proposing directions for more advanced architectures and broader market coverage in future work.
Abstract
We investigate a number of Artificial Neural Network architectures (well-known and more ``exotic'') in application to the long-term financial time-series forecasts of indexes on different global markets. The particular area of interest of this research is to examine the correlation of these indexes' behaviour in terms of Machine Learning algorithms cross-training. Would training an algorithm on an index from one global market produce similar or even better accuracy when such a model is applied for predicting another index from a different market? The demonstrated predominately positive answer to this question is another argument in favour of the long-debated Efficient Market Hypothesis of Eugene Fama.