Using N-BEATS for Interpretable Time-Series Forecasting with Market Data

Summary

The article explains how N-BEATS forecasts time series and uses constrained model blocks to represent components such as trend and seasonality. Each block maps an input sequence through a lower-dimensional representation and produces both a forecast and a reconstruction of the past. It also introduces covariates, distinguishing variables known in advance from those whose future values are unavailable, a distinction that matters when forecasting multiple bars.

The implementation discussion applies the model to MetaTrader gold data at a 15-minute interval, using closing prices as the target and discussing open, high, and low prices as possible inputs. The article describes training, validation, learning-rate selection, and plotting model interpretations, but the supplied text gives no quantitative forecast results or trading performance. It notes limitations in the forecasting library’s N-BEATS covariate support and distinguishes related external-variable models as outside the article’s scope. Interpretable decomposition can help inspect model structure, but does not establish forecast accuracy or trading value.

Key ideas

• N-BEATS uses stacked blocks to generate forecasts and backcasts from time-series inputs.
• Constraining block outputs with basis functions can support trend and seasonal decomposition.
• Covariates must be classified by whether future values are known at prediction time.
• Open, high, and low prices may be unknown covariates for a multi-bar forecast.
• The example applies the model to gold price data, but the provided material reports no trading results.

Tags

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