Wednesday, May 6, 2015

How to Forecast Gasoline Prices

From Barron's Commodities Corner May 1, 2015:
Researchers claim they’ve found the best way to forecast prices at the pump. They say it points to higher fuel costs.
Bloomberg Commodity Index
For many investors, a coin toss may seem like the most reliable way to forecast U.S. gasoline prices. But researchers say they’ve developed a system to tilt the odds.

Unfortunately, their statistical technique indicates that gas prices are likely to march steadily higher over the next 18 months.

The academic researchers looked at prices that drivers pay at the pump and found that a big part of the plunge in the second half of 2014 wasn’t as surprising as some analysts thought. Up to 39% of the decline, to be precise, was predictable as of June 2014, according to a January-dated working paper released by the Frankfurt-based Center for Financial Studies.

“The change in the gasoline price in the past has been considered essentially unforecastable, based on publicly available information,” states the paper, with the title “Inside the Crystal Ball: New Approaches to Predicting the Gasoline Price at the Pump” by Christiane Baumeister of the Bank of Canada, Lutz Kilian of the University of Michigan, and Thomas K. Lee of the U.S. Energy Information Administration. “Our findings challenge this conventional wisdom,” says the paper, which is available at the University of Michigan Website (umich.edu) under Kilian’s papers.

Over time, prices at the pump and futures prices tend to move in the same direction. So, in theory at least, investors could use the same price trends to trade.

ALTHOUGH THE RESEARCHERS tested many forecasting methods, the most accurate single model for predicting prices turned out to be the simplest, despite a forbidding name for nonstatisticians. It uses a method called vector autoregression, in this case with only two inputs: retail gas prices and the benchmark Brent crude oil price, both adjusted for inflation. Very basically, this method employs historical prices of the two commodities to forecast their future prices.

Tested against data back to 1992, the model produced statistically robust forecast results that outperformed the coin toss out to 18 months. On average, the individual forecasts were closer to actual outcomes than other models, the report shows. Think of the forecasts as target shooting with a rifle: Accuracy is important, but so is a tight grouping of shots. In this case, the grouping of forecasts for gas prices was tighter than other models.
The researchers also found that the U.S. Energy Information Administration’s own forecast, which it makes public, was directionally correct. The EIA predicted the price of gasoline would average $3.31 a gallon last December, down from $3.75 in May. The actual average was $2.54. The major problem with the public forecast is that it only extends out 12 months. The researchers’ model has a longer timeline....MORE