Crude Oil Inventory

Each week (usually Wednesdays) the U.S. Energy Information Administration publish the number of barrels of crude oil held in inventory by commercial firms during the past week. The figure is eagerly anticipated by energy traders as it is interpreted as an indicator of supply. Thus, an increase (decrease) in the figure is typically associated with a bearish (bullish) sentiment toward crude oil.


The following chart shows the trading action around the Jan 30th 2013 release over a three minute window starting one minute prior to the release, which occurs at 15.30pm GMT, and ending two minutes following. A green dot indicates an up tick, a red dot indicates a down tick, and a white dot is displayed when a trade occurs at the same prices.

The next figure shows the bid ask spread at 50 millisecond intervals over the same time horizon.

And the following figure shows the midpoint, which is defined as an average of the bid and ask prices.

Very little trading takes place over the one minute period prior to the release. The next plot shows the trading volume aggregated over 5 second intervals.

To estimate volatility I aggregate 50 millisecond returns into 5 second intervals, where return is defined as the change in the log of the midpoint. The volatility graph below shows basis points on the y-axis. As expected there is a spike in volatility on the release.

Each week market analysts forecast the inventory figure. The following plot shows the consensus forecast over 2012 versus the actual figure.

As the following plot highlights, there is usually a significant differential between the consensus forecast and the actual outcome. The differential, or forecast error, is measured in millions of barrels of crude oil.

In the next plot I show the cumulative log returns of the midpoint for each announcement over a 30 second window starting 10 seconds prior to the release. As expected there is little activity prior to the announcement and a significant volatility regime switch at 15.30pm.

The volatility switch is better illustrated in the following figure, which shows the absolute 50 millisecond returns.

Given that the market typically prices in the consensus forecast prior to the announcement, one might expect there to be a positive correlation between forecast error and post release volatility.

To test this hypothesis I investigate the correlation between the consensus forecast error and post announcement volatility. The volatility estimate is scaled to a one hour horizon. The figure below suggests that an unexpected increase in supply has a bigger impact on volatility than an unexpected decrease.

The final figure shows the relationship between the volatility estimate and the absolute forecast error. I find a significant linear correlation of 45%.

I plan to post more analysis on this topic soon! Thanks for reading!