The Past as Prologue: A New Approach to Forecasting

26 Pages Posted: 17 Aug 2020 Last revised: 12 Mar 2021

See all articles by Megan Czasonis

Megan Czasonis

State Street Corporate

Mark Kritzman

Windham Capital Management

David Turkington

State Street Associates

Date Written: August 10, 2020


It is common practice to forecast social, political, and economic outcomes by polling people about their intentions. This approach is direct, but it can be unreliable in settings where it is hard to identify a representative sample, or where subjects have an incentive to conceal their true intentions or beliefs. The authors propose that, as a substitute or a supplement, forecasters use historical outcomes to predict future ones. The relevance of historical events, however, is not guaranteed. The authors apply a novel technique called Partial Sample Regression to identify, in a mathematically precise way, the subset of events that are most relevant to the present. The outcomes of those events are then weighted by their relevance and averaged to give a prediction for the future. The authors illustrate their technique by showing that it correctly predicted the winner of the last six U.S. presidential elections based only on the political, geopolitical, and economic circumstances of the election year.

Keywords: Informativeness, Mahalanobis distance, Partial sample regression, Relevance, Similarity

JEL Classification: C00, C01, C02, C10, C13, C50

Suggested Citation

Czasonis, Megan and Kritzman, Mark and Turkington, David, The Past as Prologue: A New Approach to Forecasting (August 10, 2020). MIT Sloan Research Paper No. 6166-20, Available at SSRN: or

Megan Czasonis (Contact Author)

State Street Corporate ( email )

1 Lincoln Street
Boston, MA 02111
United States

Mark Kritzman

Windham Capital Management ( email )

800 Boylston Street
30th Floor
Boston, MA 02199
United States
6174193900 (Phone)
6172365034 (Fax)

David Turkington

State Street Associates ( email )

United States

Do you have a job opening that you would like to promote on SSRN?

Paper statistics

Abstract Views
PlumX Metrics