• No results found

per cent and that for 966-7 to over 2 per cent.

extra i observations should be excluded from both the policy "on"

1 per cent and that for 966-7 to over 2 per cent.

The study by the NBPI (1968) uses annual data and distinguishes

between tight and moderate policy; 1965 being classed as the

latter and 1961-2 and 1966 as the former. The estimated policy

effects are not statistically significant but the coefficients are

almost identical at around 1 per cent. The study by Parkin et a l .

(1976) includes only two dummy variables; the first covering the

pay pause of 1961-2 and the second the period 1966(3) - 1967(2).

Neither policy dummy proves statistically significant, the estimate

for the first policy phase being positive whilst the second phase

has a coefficient of the expected negative sign which implies a

reduction in wage inflation of just over 1 per cent.

All the studies cited so far use the Phillips curve as the

basic form of wage equation. However, despite regular resuscitation,

confidence in the Phillips curve as an empirical description of

wage behaviour was low in the 1970s and the study by Henry et a l .

(1976) turned to the real wage resistance model. Henry et a l . (1976)

also use the dummy variable approach and extend the estimation period

to 1974. They distinguish between two periods of policy after

1961, the wage-freezes of 1966 (1966(3) - 1967(2)) and of 1972

(1972(4) - 1973(4)). Only the former policy period emerges as

statistically significant with an effect of over 1 per cent.

Sheriff (1977) uses a neo-classical model of wages whereby

wages depend on output, employment, prices and time. The model is

applied to manufacturing industry using the intercept dummy

approach. However, Sheriff goes one step further than many of

the other studies by identifying fourteen separate dummies, of

also specified and the model estimated simultaneously over the

period 1959(1) to 1973(4). Two of the announcement dumnies are

significant - those for the freeze of 1966(3) and for the freeze

of 1972(4). However, whilst the coefficient on the former is

negatively signed that on the latter is positive. Sheriff explains

this result by differences in the timing of the policy announcement

within the quarter. However, Sheriff's definition of the announce­

ment effect is perhaps ambiguous. Whereas one interpretation

of the announcement effect would be that effect on wages which

occurs between declaring the intention to apply the policy and

actual commencement of the policy, an alternative interpretation used

by Sheriff is whether the impact effect on wages differs from the

continuous effect. Both the significant announcement effects

have comparable magnitudes of response, namely 3 per cent. Of the

seven continuous dummies, three are statistically significant:

those covering the period 1962(3) — 1964(4); 1966(4) - 1967(1); and

1968(3) - 1969(4). Two of the announcement dummies are almost

identical in magnitude to the succeeding continuous dummies (1961(3)

and 1966(3)). It is difficult to see why the period between 1967(2)

and 1968(2) was excluded from the analysis since a tighter policy

was clearly in operation then than in the following eighteen months.

Although the Sheriff approach is less restrictive in its coverage

of policy it does lead to the problem of multiplicity of dummies

referred to earlier. For example. Sheriff uses three separate

dunsny variables in 1973 to cover only four observations with the

consequent problem that the implied policy estimates will also

include random errors.

Henry and Ormerod (1978) also use a real wage resistance

model and include observations from 1961 to 1977(2). However the

2.24

also specified and the model estimated simultaneously over the

period 1959(1) to 1973(4). Two of the announcement dummies are

significant - those for the freeze of 1966(3) and for the freeze

of 1972(4). However, whilst the coefficient on the former is

negatively signed that on the latter is positive. Sheriff explains

this result by differences in the timing of the policy announcement

within the quarter. However, Sheriff’s definition of the announce­

ment effect is perhaps ambiguous. Whereas one interpretation

of the announcement effect would be that effect oh wages which

occurs between declaring the intention to apply the policy and

actual commencement of the policy, an alternative interpretation used

by Sheriff is whether the impact effect on wages differs from the

continuous effect. Both the significant announcement effects

have comparable magnitudes of response, namely 3 per cent. Of the

seven continuous dummies, three are statistically significant:

those covering the period 1962(3) - 1964(4); 1966(4) - 1967(1); and

1968(3) - 1969(4). Two of the announcement dummies are almost

identical in magnitude to the succeeding continuous dummies (1961(3)

and 1966(3)). It is difficult to see why the period between 1967(2)

and 1968(2) was excluded from the analysis since a tighter policy

was clearly in operation then than in the following eighteen months.

Although the Sheriff approach is less restrictive in its coverage

of policy it does lead to the problem of multiplicity of dummies

referred to earlier. For example, Sheriff uses three separate

dummy variables in 1973 to cover only four observations with the

consequent problem that the implied policy estimates will also

include random errors.

Henry and Ormerod (1978) also use a real wage resistance

model and include observations from 1961 to 1977(2). However the

2.26

Notes to Table 2.

For general methodology see Table 2.1.

* denotes statistically significant policy effect.

NBPI (1966) uses an annual model; and distinguishes between types

of policy. 1961-2 and 1966 are regarded as tight and 1965 as loose.

Smith (1968). Results quoted are for the regular dummy variable and its effects on weekly wage rates.

Sheriff (1977) uses announcement dummies and continuous dummies, see Table 1.1.

Henry and Ormerod (1978): preferred equation from Table 9.

Effects are on the rate of acceleration of wage inflation. Estimates marked t are catch-up estimates.

Sargan (1980): details are taken from an earlier version of this

paper as they are not given in the final version. Figures in brackets refer to instrumental variable estimates.

2 .27

that used by Henry et a l . (1976) and consequently the dependent

variable is defined as the rate of acceleration of wage inflation.

The policy effects estimated should therefore be interpreted in this

way, as should the policy catch-up effects which Henry and Ormerod

allow for (but see the discussion earlier in the chapter). In

addition to the standard policy and catch-up dummies Henry and

Ormerod also include a shift dummy after 1975(2) to prevent the

model from breaking down. Even so it should be noted that only three

of the thirteen coefficients in their equation are statistically

significant at the conventional level (and one of the coefficients is the

first-order serial correlation coefficient). Henry and Ormerod are

not able to distinguish between alternative hypotheses regarding this

additional shift dummy. It is not clear therefore whether it can

be attributed entirely to incomes policy. The two significant policy

episodes are the period 1967(3)-1969(2) (but not the preceding freeze)

and the catch-up variable for this period. Although many of the

values of the catch-up effects are similar to those of the relevant

policy estimates no conclusion about catch-up effects is possible

given the lack of statistical determinacy.

All the studies referred to have treated incomes policy as

exogenous. An exception is the study by Sargan (1980). He uses

both the policy on/off and the dummy variable approach and in the

latter treats the dummy variables as endogenous variables within

estimation. Sargan finds that the policy on/off approach leads to

18