• No results found

CHAPTER 4: REGRESSION ANALYSIS OF SEASONALITY AND STATE-DEPENDENCE IN PRICING

4.4
 Analysis
of
frequency
of
price
changes
using
CPI
micro­data


4.4.2 Real-time explanatory variables

There is some clear evidence of real-time state-dependence in pricing. Firstly, there is evidence reported in Table 44 that the frequency of price changes is positively and significantly associated with the level of the CPI rate of inflation. This finding is robust across the various specifications and holds for both the level of the CPI and for changes in the level of inflation. From Appendix 10.1 it is clear that this is due to the fact that the frequency of price increases is significantly and positively associated with a change in the CPI rate. There is also evidence, presented in Appendix 10.2, of a statistically significant negative real-time relationship between the frequency of price decreases and the level of inflation. This means that as the level of inflation rises, the frequency of price decreases falls. Interestingly, the finding of a positive relationship between inflation and price change frequencies is held up in the literature (such as in Klenow et al (2008) and in Gagnon (2009)), as providing evidence of synchronisation in pricing conduct, where rising inflation is associated with an increase in the frequency of price increases.

Secondly, there is evidence of a statistically significant negative real-time association between the frequency of price increases and an appreciation of the nominal effective exchange rate. As outlined in Appendix 10.1 this finding is robust across the various specifications of the model, and as such an appreciation of the currency is associated with a small decrease in the frequency of price increases. On the other hand, the frequency of price decreases is positively and significantly associated with an appreciation of the nominal effective exchange rate. Thus, there is evidence of a pass-through effect in that pricing conduct is influenced by exchange rate developments in the expected manner, although, as outlined in Appendix 10.2, this finding is not robust across the various specifications of the model. It is possible that the two effects – a decrease in the frequency of price increases and an increase in the frequency of price decreases - cancel each other out as, at the aggregate level, there is no statistically significant relationship between a currency appreciation and the overall frequency of price changes in the full specification of the model. This result is not robust across all specifications, as per Table

statistically significant decrease in the frequency of price changes associated with a currency appreciation.

Thirdly, the frequency of price changes is significantly and positively associated with the Repo rate level and with changes in the Repo rate, if these independent variables are included individually. This result is significant at the 1% level as reported in Table 44. However, where the Repo rate levels and the change in the Repo rate are included with the other independent variables in the combined specification of the model, then only the level of the Repo rate remains statistically significant. This suggests that it may be the level of the Repo rate, rather than changes to the Repo rate, that is associated positively in real-time with the frequency of price changes. From Appendix 10.1 and 10.2 respectively, it can be seen that it is both the frequency of price increases and the frequency of price decreases that are positively associated with the level of the Repo rate. The case of price increases being robust across all specifications of the model, and the case of price decreases only holding with the combined specification of the model. It is difficult to interpret the implications of this finding, as a priori an increase in the Repo rate should, due to the retardation of aggregate demand, be associated with an increase in the frequency of price decreases, and a decrease in the frequency of price increases. Perhaps, the rise in real-time in the frequency of price increases is indicative of some cost channel effects, whereby interest rate increases translate into higher prices.23 Alternatively, the real time positive association between the frequency of price increases and the level of the Repo rate, may be indicative of a causal direction where the Repo rate is increasing in response to the (relatively high) frequency of price increases, and thereafter it would be expected that the decrease in the frequency of price increases would only become apparent after some time had passed, due to lag effects. The statistically significant (at the 1% confidence interval) positive association between the frequency of price decreases and the level of the Repo rate (reported in Appendix 10.2) may be indicative of the expected monetary policy transmission mechanism effect that would associate a higher Repo rate (particularly after a lag) with reduced aggregate 







23See Barth and Ramey (2001) and Ravenna and Walsh (2006) for full discussion on cost channel effects and the impact of such effects on monetary policy transmission.

demand and a higher frequency of price decreases.24 The finding is not robust across all specifications as it only occurs in the combined version of the model.

Fourthly, the positive coefficients, in Table 44 and in Appendix 10.1 and 10.2, on the dummy for the break in the data set at 2006m3, offer a strong indication, significant at the 1% level, that the frequencies of price changes, price increases and price decreases all increase in the second part of the CPI data set. As explained, this is the likely effect of the inclusion of sales prices, which would tend to display such higher price change frequencies. This finding is robust for all specifications of the model.