An empirical testing of Basu′s `conservatism
principle′ of stock prices

Felix Brandes

Introduction:

This report will test Basu′s (1997) hypothesis that the slope coefficient and 2

R

-value from a

regression of annual earnings on annual returns is higher for negative returns than for positive

returns. A cross-section of UK firm data with 931 observations from the year 1996 will be

used as a sample to test this hypothesis. The report will start with a short discussion of the

underlying economic rationale that is thought to cause this behaviour. It will then define the

sample, regression models and variables that are used to test the hypothesis and present the

empirical results. The report will conclude with an interpretation of the results and relate these

to Basu′s findings.

-

2

-

Underlying economic rational:

Basu, in her article `the conservatism principle and the asymmetric timeliness of earnings′,

predicts that earnings is more timely in reflecting `bad news′ than `good news′ (Basu, 1997).

She identifies accounting conservatism as the source of this asymmetric behaviour (ibid). The

rational behind it is that accountants have to recognise expected losses right away for matters

of prudence, whereas they cannot recognize gains until they have `materialized′. Therefore

losses are reflected in earnings, which is subject to accounting conventions, much timelier

with the announcement, whereas gains may take several years to affect earning numbers.

However, the share price, and hence the return, is subject to valuations of stock market

participants. They do not use overly conservative predictions but use their expectations to

model an appropriate price. Good news or bad news therefore should change the share price

instantly

to the extent the news affect

expected

future earnings. Therefore, we have a

structural difference in behaviour between `bad news′ and `good news′ in relation to returns

(ibid). It follows that negative returns proxy for `bad news′ and positive returns proxy for

`good news′. As `bad news′ are reflected more quickly in accounting earnings due to

conservatism, it is expected to move much closer with negative return changes. Therefore,

conservatism leads to an asymmetric recognition of `bad news′ and `good news′, being

reflected in earnings in regard to time. For this reason a higher slope coefficient and 2

R

-value

for `bad news′ is expected than for `good news′.

Defining the sample, variables & calculations:

As my library card number ends on 69, group six and nine was deleted from the original

sample as instructed, which reduced the sample size to 931. The sample data consist of UK

firm data from the year 1996 taken from the `Datastream′ database. The calculations for the

regression variables are based on four sets of data. These are `earnings per share for 1996′,

dividends per share in 1996′, price per share at the end of 1996′ and `price per share at the

end of 1995′. All numbers are expressed in GBP pence.

Unlike most prior research, Basu models an earnings-return regression to explain earnings

behaviour (ibid). The basic earnings-return model,

Regression 1

, is the following:

X

P

- = +

R

it

it

1

0

0

it

The model uses earnings in period

t

(

X

) as the dependent variable, deflated with last years

it

share price (

P

) to reduce autocorrelation, and a return variable (

R

) as the independent

it

1

-

it

-

3

-