From VoxEU:
Recent work on forecasting oil prices raises the question of whether oil
industry analysts know something about forecasting the price of oil
that academic economists have missed. This column presents evidence that
they do, but economists know how to improve further on these
practitioners’ insights.
Petroleum products
such as gasoline and heating oil are produced by refining crude oil.
Many oil market analysts believe that the prices for these petroleum
products contain useful information about the future evolution of the
price of crude oil. In particular, changes in the product price spread –
defined as the extent to which today’s price of gasoline or heating oil
deviates from today’s price of crude oil – is widely viewed as a
predictor of changes in the price of crude oil. For example, in April
2013 Goldman Sachs cut its oil price forecast citing significant
downward pressure on product price spreads, which it interpreted as an
indication of reduced final demand for products. Likewise, in 2011
energy consultant Kent Moors predicted higher oil prices based on
widening gasoline and heating oil-price spreads.
Although energy
economists have made great strides in recent years in forecasting the
price of oil at short horizons, the forecasting ability of product
spreads has never been formally analysed to date. Our recent work asks
whether academic economists have missed something about forecasting oil
prices that oil industry analysts know. The answer is that they have,
but so have practitioners. Based on a rigorous real-time out-of-sample
evaluation of numerous oil price forecasting models, we find that not
all product spread forecasting models are useful in practice. Some
forecasting models used by oil market analysts lack a solid foundation,
but there are alternative product price spread models that greatly
improve our ability to forecast the real price of oil. We develop
forecasting models based on the gasoline-price spread that are
systematically more accurate in real time compared with conventional
no-change forecasts.
Such models work particularly well at forecast
horizons between one and two years, far beyond the short horizons for
which earlier oil-price forecasting models based on economic
fundamentals have been shown to work well. We obtain even more accurate
results with a model that allows the predictive power of gasoline price
spreads and heating oil spreads to evolve over time.
Predicting with spreads
Our study is based
on the proposition that that the price of crude oil can be expressed as a
weighted average of product prices. This proposition has a long
tradition in energy economics. For example, oil analyst Philip K.
Verleger popularized the idea that the demand for crude oil ultimately
derives from the demand for refined products, with refiners buying crude
oil only if they can generate a profit at prevailing product prices.
Our forecast analysis does not depend on this economic interpretation;
all that is required to motivate the forecasting models in question is
that the price of oil and the product prices share a common trend.
The study considers
four basic forecasting models based on spreads with futures prices as
well as spot prices for gasoline and heating oil:
- Models of individual product spreads such as the gasoline-price spread or the heating oil price spread.
- Models based on weighted product spreads.
- Models based on the crack spread, and
- Equal-weighted forecast combinations of gasoline spread and heating oil-spread models.
The evaluation
period extends from early 1992 until September 2012. The study evaluates
the out-of-sample accuracy of each of the forecasting models in terms
of the recursive mean-squared prediction error (MSPE) relative to the
no-change forecast and based on their ability to predict the direction
of change in the real price of oil.
We find that not all product spread models are useful for out-of-sample
forecasting, but some models perform well. The best single-spread
forecasting model is a model based on the gasoline spot spread alone
which yields MSPE reductions as large as 15% and directional accuracy as
high as 63% at the two-year horizon. Heating oil spot spreads are far
less accurate predictors than gasoline spot spreads. Weighted product
spread models are never more accurate than gasoline spread models.
Perhaps surprisingly, there is no evidence of forecasting models based
on the commonly cited 3:2:1 crack spread having out-of-sample
forecasting ability....
MUCH MORE
