News & Events

UI Research Suggests That Financial Models Can Benefit from Lessons in Meteorology

A group of University of Iowa researchers thinks that economists and investors might find it useful to take a cue from meteorologists.

The researchers—economics and finance professors in the Tippie College of Business—are looking at a new way of using economic and financial models to more accurately predict stock returns. In the postmortem of the Crash of 2008, economists were reminded that economic models are fallible when it comes to predicting the future, especially when a black swan starts swimming in the risk pool. The hope is that with a more accurate tool, the chance of model failures will diminish.

Rather than using a single model to make economic or financial forecasts, the Tippie researchers are testing the theory that researchers and investors run several different models when they formulate an economic forecast. Their early work suggests that these model pools are better at predicting future trends than any one single model.

“At a subconscious level, it seems obvious that you should not rely on any one model to make financial decisions, but people tend to get carried away with the elegance of a model,” said Ashish Tiwari, associate professor of finance in the Tippie College of Business. “The crisis reinforced the well-known fact that all models are only approximations of reality.”

Tippie researchers like Tiwari and economics professor Gene Savin are looking at how to make models more useful in practice. Building on the work of a former Tippie economics colleague, John Geweke, they have focused on the concept of using a pool of models to devise an economic forecast that is not unlike how a weather forecast is put together, Tiwari said. Meteorologists typically don’t rely on a single model when they develop their weather forecasts. Instead, they run several models that incorporate different types of climatologic information and then synthesize the results of each to determine if it’s going to be rainy or sunny tomorrow. While a single model might be fine when the weather is stable, it would leave too many holes when, say, trying to predict the path of a hurricane.

“That’s why when you see the weather reporter standing on the beach as the hurricane approaches, they tell you it’s expected to make landfall anywhere between Pensacola or Pascagoula,” he said. “Different models predict different paths.”

Taking that inspiration from The Weather Channel, Tiwari, Savin, and Tippie finance doctoral student Michael O’Doherty recently put together a series of model pools to see how they would predict returns on stock portfolios between 1932 and 2008. After comparing the prediction to what actually happened, they found that a two-model pool led to more accurate predictions than any one model, and a three-model pool was even more accurate.

The reason, they explained, is that each model in the pool covers for the weaknesses of others, with the result that the pool captures more information than any single model. The initial set of models used are the CAPM, the Fama-French model, and the Carhart model, what Tiwari calls “the workhorse models” of empirical asset pricing. The researchers ran each model monthly to predict portfolio returns during the test period, and then averaged the three predictions on a weighted basis. Two more models were eventually added to bring the size of the pool to five models.

“Adding more models to the pool generally improves the accuracy of the predictions, provided the new models contain unique information,” Savin said.

The researchers say that model pooling can potentially be used to produce more accurate predictions for a wide range of economic applications, whether it’s real estate, unemployment, or GDP. As a part of their follow-up research, Tiwari, Savin, and O’Doherty are currently looking at model pools to improve upon the benchmarks used to evaluate the performance of mutual fund and hedge fund managers.

Their research on stock portfolio returns is presented in a paper titled, “Modeling the Cross Section of Stock Returns: A Model Pooling Approach.”


Return to top of page