Discussing the article: "Neural Networks in Trading: Lightweight Models for Time Series Forecasting"
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Check out the new article: Neural Networks in Trading: Lightweight Models for Time Series Forecasting.
Lightweight time series forecasting models achieve high performance using a minimum number of parameters. This, in turn, reduces the consumption of computing resources and speeds up decision-making. Despite being lightweight, such models achieve forecast quality comparable to more complex ones.
The foundation of precise long-term time series forecasting lies in the inherent periodicity and trends present in the data. Additionally, it has long been observed that the price movements of currency pairs are closely related to specific trading sessions. For instance, if a time series of daily sequences is discretized at a specific time of day, each subsequence exhibits similar or sequential trends. In this case, the periodicity and trend of the original sequence are decomposed and transformed. Periodic patterns are converted into inter-subsequence dynamics, while trend patterns are reinterpreted as intra-subsequence characteristics. This decomposition opens new avenues for developing lightweight models for long-term time series forecasting, an approach explored in the paper "SparseTSF: Modeling Long-term Time Series Forecasting with 1k Parameters".
In their work, the authors investigate, perhaps for the first time, how periodicity and data decomposition can be leveraged to build specialized lightweight time series forecasting models. This approach enables them to propose SparseTSF, an extremely lightweight model for long-term time series forecasting.
The authors present a technical method for inter-period sparse forecasting. First, the input data is divided into constant-periodicity sequences. Then prediction is performed for each downsampled subsequence. Thus, the original problem of forecasting time series is simplified to the problem of forecasting the interperiod trend.
Author: Dmitriy Gizlyk