Compustat Unrestated Quarterly Data

The Unrestated Quarterly data items have been retrieved from "Historical" Compustat Data, using prior releases of then-Current Compustat. More precisely, the Unrestated Quarterly data items are defined as the data values when each company’s quarterly data was first presented by Compustat as "final" prior to any subsequent quarterly data Restatements, for all companies in the Compustat database (both current and research, domestic and Canadian).

The magnitude of the difference between Restated and Unrestated Quarterly data values varies from trivial to enormously significant. For any given data Restatement, these differences vary independently on each of three dimensions: (1) from company to company; (2) from time period to time period; and (3) from data item to data item.

Accordingly, trying to predict or statistically filter to correct for quarterly restatements is a random walk through a mine field. One may see little quarterly restatement difference for one factor or company or quarter's results yet find for another quarter that the magnitude of the difference totally changes the set or rankings of companies and the identified outliers for one or more factors.

Percentage of Companies With Restatements

A significant amount of Compustat data is restated, this is because a restatement does not only affect the quarter the restatement happened, it also can affect figures for previous quarters.

The chart below shows the percentage of companies with restatements over time for net income or component data in the S&P 500, next and second thousand largest companies by market capitalization and for all traded companies.

At any point in time, the S&P 500 had at least 35% of its companies' release restatements!

Data Restatements

The chart below illustrates the percentage of companies, by market capitalization, that have anywhere from a 5 to 25 percent difference in what was originally reported and what is the current value in the database for net income or some component data. Notice over time roughly 35% of the S&P500 had restatements greater than 25% of the original finalized value reported.

Accordingly, Backtesting with Current Compustat data risks using a seriously discontinuous time series of substantially random quarterly data values for a significant number of companies due to the overwriting of Compustat data with Restatements.

The severity of using restated data in a backtest can be seen in the below graph; a classic example on how one can be fooled, see the huge jump in the sales-to-price ratio of Exxon Mobil prior to the merger in the 4th quarter of 1999 between Exxon Mobil because the values were restated to show what the combined entity would have looked like prior to the merger.

So, in actuality, the sales-to-price ratio continues to fall, but the restated Sales values of Exxon and Mobil compared to Exxon's monthly closing price, created the fictitious spike potentially placing this stock in your portfolio.