Introducing BISTRO: a foundational model for unconditional and conditional forecasting of macroeconomic time series
Summary
Focus
The ever-growing capabilities of large language models (LLMs) have opened up the prospect that a similarly flexible approach would support time series forecasting. Just as LLMs do well in guessing the next word within the broader context of the sentences that precede it, foundational time series models could do well in guessing the next realisation of a macroeconomic time series within the broader context of what has been happening in the economy. Hence, rather than having to build a bespoke model for a particular task, the same foundational model could be deployed for a wide variety of different tasks.
Contribution
We introduce the BIS Time-series Regression Oracle (BISTRO). BISTRO builds on the machinery underlying LLMs and applies it to the world of economic time series. BISTRO comes with detailed instructions and pre-compiled scripts on how to operate it, available at https://github.com/bis-med-it/bistro. To facilitate replication and practical use, these scripts can be run in Google Colab, allowing users to upload their own data set and generate baseline and conditional forecasts with BISTRO through a guided workflow.
Findings
BISTRO can assist economists in their forecasting and scenario analyses. For example, a researcher can produce a generic baseline forecast for, say, inflation and then evaluate how conditioning on different explanatory variables (and different assumptions for their evolution) modifies the baseline. Hence, BISTRO constitutes a low-cost and easy-to-use forecasting tool that performs well compared with traditional econometric benchmarks.
Abstract
This article introduces the BIS Time-series Regression Oracle (BISTRO), a general purpose time series model for macroeconomic forecasting. Its edge over traditional econometric approaches lies in its ability to deal with generic unconditional and conditional forecasting tasks, without requiring to adjust the model to the macroe conomic tasks being tackled. Building on the transformer architecture underlying LLMs, BISTRO is fine-tuned on the large repository of macroeconomic data main tained at the BIS. We show that BISTRO provides reliable unconditional forecasts for key macroeconomic aggregates and illustrate how using it for conditional fore casting can help unveiling patterns of nonlinearity in the data.
JEL classification: C32, C45, C55, C87
Keywords: forecasting, scenarios, large language models
