Findings and analysis

Benchmark results

We use the estimations for the indicator with five values (from -2 to 2) as the benchmark results, presented in Table 4.⁶ In the basic specification with only growth and inflation when the official targets are used (Column 1 of Table 4), the coefficients on economic growth and inflation are highly significant, demonstrating the importance that the PBoC attaches to its legal mandate of maintaining stable economic growth and low inflation. Among the two targets, the policy response for economic growth is stronger than that for inflation.⁷ This suggests that the PBoC shows stronger reaction to each percentage-point deviation in economic growth from the official target than that for inflation. The stronger policy response to economic growth may suggest that real GDP growth is taken as a good summary indicator of economic developments, and monetary policy is believed to have a strong impact on it.

6 This and all other specifications have been estimated by both the probit and logit models, and the results are very similar. All the results presented are from the logit model estimation, which offers slightly better fit of the data.

Monetary policy also influences inflation. However, some volatile items in the consumer price index such as food may be influenced by non-monetary factors such as weather conditions, while utility prices are still subject to administrative controls in China. Under the circumstances, other means have been used to complement monetary policy in controlling inflation, and monetary response to inflation movements is thus milder. For example, in early 2008 when there were heightened concerns about inflation, measures were announced by the National Development and Reform Commission and State Council to stabilise prices.

The model has reasonable explanatory power. For 69.7% of the 33 observations, the implied monetary stance based on the estimated model correctly signals the actual stance.

Among the other economic objectives, monetary and credit growth, and unemployment, are largely not statistically significant, and often carry a wrong sign (Columns 2 - 4 of Table 4). The PBoC does not appear to directly react to changes in monetary conditions, or the employment situation.⁸

The results are broadly similar to the estimation using historical averages as targets (Columns 5 - 7 of Table 4). Both economic growth and inflation are important determinants of monetary stance, with growth being a more significant factor. The basic specification with only growth and inflation improves on the goodness of fit compared with that using the official targets. Monetary stance is correctly signalled in 75.8% of the cases. It is better at accurately signalling tightening stance, being correct in 86.3% of the time (Table 5). The success rate drops to 62.5% for signalling loosening stance, and one third for neutral stance.

The better predictive power of using historical averages in the estimation model may suggest that a rule of thumb in policymaking using historical averages may be a better description of the PBoC’s targets than the annual official announcements. This may be explained by the way official targets are set which is discussed earlier. For

7 The size of policy response to one target is calculated as the coefficient on the target multiplied by the standard deviation of the target.

8 It is possible the estimation is affected by poor unemployment data. They only take into account urban residents who are registered as being unemployed at local employment service agencies. Given the limited coverage, the statistics may not be representative of developments in the labour market.

China Economic Issues – Number 1/10, January 2010

example, with the Central Government’s intention to regulate the growth impulse on the part of local governments, the growth target is typically set conservatively. As the outturns of growth almost always exceed the official targets (with the exception of 2008 Q4 – 2009 Q2 amidst the severe global recession), the latter may not be considered to be binding in policymaking. On the other hand, historical averages may be perceived to be more representative of the underlying growth potential of the economy, and thus be given more attention.

In contrast, the results from using the Hodrick-Prescott filtered trend growth as the target are much weaker (Columns 8 - 10 of Table 4). The coefficient on the output gap is not always highly significant, and that of the CPI fails to have any significance.

This perhaps is not surprising. As observed earlier, deviations of economic indicators from the Hodrick-Prescott filtered trend do not capture China’s economic cycles well.

Indeed, this may explain our general finding of stronger policy reaction to GDP growth which differs from that of He and Pauwels (2008). They show that the PBoC reacts more strongly towards deviations of inflation from the target than the output gap, where the output gap is derived from applying the Hodrick-Prescott filter.

Robustness check

We also undertake the estimation for the other two monetary stance indicators, one with seven values and the other three values (Tables 6 and 7). The major results from the estimation of the 5-value indicator are broadly born out by those for the 7-value and 3-value indicators. Importantly, real GDP growth and inflation are found to be the most important policy objectives, with a strong policy response to real GDP growth. Other policy objectives such as monetary and credit growth, and unemployment do not have significant impacts on monetary policy. Among the models using different sets of targets, the ones with historical averages as the target have the best fit for the data, while those with the Hodrick-Prescott filtered trend growth are the poorest.

Also we have experimented with eight and 10 years as the window for calculating historical averages as monetary targets in estimating monetary reaction functions.

Again, the results are broadly similar as above. Most notably, using historical averages stays the best way of describing the PBoC’s policy targets.

While the major findings are robust, the explanatory power varies among the three different monetary stance indicators. The explanatory and predictive power rises as the number of values in an index declines, probably reflecting difficulties in differentiating the intensity of monetary stance. The model of the 7-value indicator has the lowest predictive power, making correct predictions in 59.4% of the cases in the benchmark specification with only growth and inflation included as policy objectives, and historical averages as the target set. That rises to 75.8% for the model of the 5-value indicator, and further to over 90.9% for the model of the 3-value indicator. Most notably, the model of the 3-value indicator is 100% correct in predicting tightening stance (Table 8). Also similar to the predictions of the model using the 5-value indicator, the success rate is lower for predicting loosening stance (87.5 percent), and the lowest for neutral stance (one third).