The M score was created by Professor Messod Beneish. In many ways it is similar to the Altman Z score, but optimised to detect earnings manipulation rather than bankruptcy. The orginal full paper can be found here www.bauer.uh.edu/~swhisenant/
Beneish used all the companies in the Compustat database between 1982-1992. The M score is based on a combination of the following eight different indices:
DSRI = days' sales in receivables index (measured as the ratio of days' sales in receivables in year t to year t-1). A large increase in DSR could be indicative of revenue inflation.
GMI = gross margin index (measured as the ratio of gross margin in year t-1 to gross margin in year t). Gross margin has deteriorated when this index is above 1. A firm with poorer prospects is more likely to manipulate earnings.
AQI = asset quality index (asset quality is measured as the ratio of noncurrent asses other than plant, property and equipment to total assets). AQI is the ratio of asset quality in year to year t-1.
SGI = sales growth index (ratio of sales in year t to sales in year t-1). Sales growth is not itself a measure of manipulation. However, growth companies are likely to find themselves under pressure to manipulate in order to keep up appearances.
DEPI = depreciation index (measured as the ratio of the rate of depreciation in year t-1 to the corresponding rate in year t). A DEPI>1 indicates that assets are being depreciated at a slower rate. This suggests that the firm might be revising useful asset life assumptions upwards, or adopting a new method that is income friendly.
SGAI = sales, general and administrative expenses index (the ratio of SGA expenses in year t relative to year t -1).
LVGI = leverage index (the ratio of total debt to total assets in year t relative to yeat t-1). An LVGI >1 indicates an increase in leverage
TATA - total accruals to total assets (total accruals calculated as the change in working capital accounts other than cash less depreciation).
These eight variables are then weighted together according to the following:
M = -4.84+0.92*DSRI+0.528*GMI+0.404*AQI+0.892*SGI+0.115*DEPI-0.172*SGAI+4.679*TATA-0.327*LVGI
A score greater than -2.22 indicates a strong likelihood of a firm being a manipulator. In his out of sample tests, Beneish found that he could correctly identify 76% of manipulators, whilst only incorrectly identifying 17.5% of non-manipulators.
When I ran M score on the non-financial companies in the S&P500 at the end of October 2003, some 57 stocks showed a M score greater than this critical threshold. I have recently updated their performance. Since October 2003, the S&P500 is up some 38%. An equally weighted basket of the stocks identified as potential manipulators is up only 21%, a 16% underperformance.
The median return on the identified stocks is just 12% (a 26% underpeformance of the S&P500). Some 73% of the stocks uncovered as potential manipulators had returns below the market, 20% had negative absolute returns!
As I was thinking about this issue a new paper by Prof. Beneish landed in my inbox (
In this new paper Beneish explores the use of the M score as a stock selection technique. Beneish examines portfolio deciles based around his M score over the period 1993-2003 with annual rebalancing done four months after the financial year end.
The results are impressive. When using market and size adjusted returns the M score strategy generates a hedged return of nearly 14% p.a. Using the Fama and French 3 factor model (market, size and style adjusted) , the stocks with the worst M scores show a -12% return, whilst the stocks with the best M scores show a 4% return, generating a long/short return of 16% (of course).
In the last year I have had many conversations with traditional long only fund managers who were setting up internal hedge funds or 130/30 style products. They were often surprised when I suggested that the shorts were not just the opposite of their longs. Shorting requires a different discipline. Perhaps using the M score might offer up a short list of potential candidates.
With earnings at a cyclical peak, finding out who has been fudging their numbers could be a particularly useful pursuit. Sadly, I don't have the data to run this screen from here, but perhaps a call to your favourite quant analyst might be in order (if that happens to be Andy Lapthorne, then you are out of luck as I know he, like me, is enjoying his garden at the moment).