Project setup

We consider the following setup.

1. Suppose we have a set of time series. In general, the set can contain from one to a relatively large number of series (say, tens of thousands).

2. For each time series we want to store actuals and to calculate and store forecasts. In particular, it is needed to store out-of-sample forecasts produced from different origins (rolling-origin forecasts) and with different horizons and, perhaps, using different methods. We also may want not only to store point forecasts, but prediction intervals (PIs), density forecasts, and additional information related to forecasting process (such as model coefficients, reasons for judgmental adjustments, etc.).

3. We assume that both actuals and forecasts may be frequently updated as new data becomes available.

Given the above settings, we need to have a convenient means to store and access (and, perhaps, to distribute or exchange) forecasting data including actuals, forecasts, and the related information. We would like to find a means that would be fast, cross-platform, easy to learn and to implement.

Eventually, the storage of forecasting data is needed to perform adequate out-of-sample evaluation of forecasting accuracy. In the case of forecasting competitions, a well-defined approach to store forecasting data should enable a credible approach for forecasting accuracy comparisons.

To cite this website, please use the following reference:

Sai, C., Davydenko, A., & Shcherbakov, M. (date). The Forvision Project. Retrieved from https://forvis.github.io/

_{© 2020 Sai, C., Davydenko, A., & Shcherbakov, M. All Rights Reserved. All rights reserved. Short sections of text, not exceed two paragraphs, may be quoted without explicit permission, provided that full acknowledgement is given.}