Monday, March 20, 2006

Leading indicators changes and statistical imputation 

The latest report on leading economic indicators includes some data revisions that are muddying the results. Besides the 0.3% decline in indicators for February, the reading for January was revised to 0.5% from 1.1%. While many will make a big deal out of this, it comes from downward revisions in two numbers which the Conference Board estimates, so that it can report out the data faster. That is, the Conference Board uses a statistical model to estimate the values in LEI for which they don't have actual data to move the series forward in the economic news cycle. At the bottom of page 3 of the report, they say it is an autoregressive model. We therefore know that an autoregressive model -- meaning that the series is forecasted as a weighted average of past observations -- did not estimate capital goods new (where they made an error more than 18%!) and for the personal consumption expenditures deflator (the price index used to create real M2 money supply, and a number the Fed is said to focus on for its measure of inflation in the economy.) Those two changes alone account for all but 0.04 of the revision to January LEI.

It's really then not news. The revision is not because of new data but due to a mistake in projecting the missing components in LEI. And since we could observe the Conference Board's errors in advance of them happening, the reaction to this report should be neutral or even slight positive.