Profit persistence in the ‘very’ long run: evidence from survivors and exiters

Abstract

One of the main shortcomings of the profit persistence literature is the fact that it looks only at surviving companies. This paper uses a unique dataset to analyse profit persistence in two different samples of stationary series: 85 surviving US companies from 1950–1999 and 72 exiters. While the exiters perform more competitively than the survivors there is still significant evidence for profit persistence in both samples. Concentration and growth of the industry as well as size and volatility of profits seem to play an important role in explaining persistence.

Acknowledgements

The author is grateful to Dennis C. Mueller for his ideas and his support and wishes to thank Elizabeth Raab for the spell check of the paper.

Notes

Average US Imports/GDP in the period 1971–2000 were 200% higher than in the period 1950–1970. Source: Bureau of Economic Analysis, http://www.bea.doc.gov/.

For a comprehensive description of the methodology see Mueller (1986 and 1990).

In order to make the profit measure independent of the source of funds used to create total assets, interest expenditures of individual companies should have been added to income before dividing by total assets. Even if annual interest rates data are available for the US from many sources as well as including those from 1950s, interest expenditures of individual companies are very often missing, especially for the years 1950–1977. Because of this reason, the afore mentioned variable could not be taken into account in this study. Additionally to the present study, a sensitivity analysis has been done for the period 1980–1999 when interest data was available and the results using interest were not significantly different from the ones without interest. Therefore the author concludes that the results reported would not have been significantly different if interest expenditures would have been taken into account.

The economy-wide measure is the median of the profit of a sample consisting of more than 175 000 observations and more than 15 500 companies. The number of annual observations is at least 677 and at most 10 710. Note that using the sample mean (or median) might be misleading. The profits of the sample studied might be not abnormal with respect to the own sample average but might be well above (or below) the economy average (or median).

Note that the specification given by Equation 2 can be justified theoretically as a reduced form of a two-equation system were profits are assumed to depend on the threat of entry in the market, and the threat is assumed to depend on the profits observed in the last period (see Geroski, 1998).

For a definition of AIC and SBC see for example SAS/ETS User's Guide, Version 8, Volume 2 page 1482.

To see why it might not be appropriate to use the criteria of highest R ² or even the criteria of highest adjusted R ² when deciding among different models, see Greene (1993), page 244.

The sample ends in 1999, as this the year for which data are available for all companies of the sample. Although Compustat provides information for some firms until 2003, it was decided to use a balanced sample with the same number of observations for each company.

As described in Section II, profits were defined as income over total assets. Comprehensive description of the two variables are available from Compustat (data definitions).

Available: http://www.wooster.edu/economics/archive/indconc.html

Available: http://www.census.gov/

Available: http://www.loglink.com/sic.asp

As mentioned before, cases where |λ _i | > 1 are not suited for the present analysis given the explosive nature of their dynamics.

The results for the full sample are available from the author upon request.

Neither the exiters series that were nonstationary due to the Dickey Fuller test, nor the whole panel of exiters were stationary due to the Im et al. (1997) test.

The economically plausible range is rather

.

Note that if you would have for example, just two companies, one with a very high but positive

and a second one with a very high but negative

then the mean

would be close to zero implying perfect competitiveness when in fact both companies have a high degree of persistence.

See Mueller (1986), page 22. Note that comparison with Mueller (1986) is slightly more suitable than with Mueller (1990) since Mueller (1986) also uses normalized profits, as the present study, while Mueller (1990) only uses firm i's return on capital.

The estimation was done in an unrestricted setting, that is lambda was not conditioned to be in the interval (−1, 1).

A similar frequency distribution for the persistence coefficient

is Table 11 (Appendix).

Visual inspection of the data on exiters confirms this claim.

General estimates of long-run projected profit rates and speed of adjustment are in Table 10 (Appendix).

All the means are significantly different from zero.

See for example Mueller (1981), Melicher and Rush (1974) and Lynch (1971).

In order to test the hypothesis for each company simultaneously N dummies for the coefficients

and N dummies for the intercepts

were constructed.

The Wald Test was employed. For more information on the theoretical properties of the Wald test see Phillips and Park (1988).

The fact that inferences made about firm performance using data on survivors is often robust to data on non-survivors was also noticed in Geroski (1998).

Note that looking for industry level patterns in the estimated

and

imposes a unique

and

level for each industry assuming that the firms in an industry are all alike. The dummy approach used here allows for firm level differences as the ‘new learning’ emphasizes.

See for example Yurtoglu (2004) and his references.

All the results for CR4-50 were similar.

A similar result was obtained by Kambhampati (1995).