The movements of the PBC's policy variables reflect more than policy changes. They incorporate large endogenous responsesof monetary policy to current and expected economic developments. This section presents a simple framework to illustrate theidentification problem inherent in the estimation of the effects of monetary policy. It further shows why estimates withconventional measures of monetary policy are likely to give rise to omitted-variable bias.Suppose output is affected by the monetary policy variable ∆m (for example, conventional monetary policy measures such as amonetary aggregate, the money market rate, or other policy indicator) and some other factors:ΔYt ¼ α0 þ α1Δmt þ α2Zt þ α3Et þ ∈t; ð1Þwhere ∆Yt is the first difference of logarithm of real GDP. The vector, Z, summarizes many other factors that affect real growth,such as government spending, supply shocks and foreign demand shocks, while the vector, E, includes expectations about futuredevelopments. For simplicity, the lagged terms of variables are ignored.The PBC reacts to output and sets policy to achieve economic goals. When doing so, it watches a range of variables, includingoutput growth and variables in the Z and E vector. Its policy reaction function can be written as:Δmt ¼ β0 þ β1ΔYt þ β2Zt þ β3Et þ δt; ð2ÞRewriting Eq. (2) yields:ΔYt ¼ −β0β1! " þ1β1Δmt− β2β1Zt− β3β1Et− 1β1δt: ð3ÞThe rearranged policy reaction function, Eq. (3), appears to have the same list of right-hand variables as Eq. (1). We hence donot know what a simple regression of output on those variables tells us. Is it the effect of monetary policy on output, or the PBC'sreaction function? Most likely, it is the mixture of these two. This is known as the identification problem. Only exogenousshocks to the monetary policy variable – δt in Eq. (2), the monetary policy movements that are independent of output growthand other factors in the vector Z and E that affect output growth – can be used to identify the true effects of monetary policy onoutput.Indeed, many papers address the identification problem and measure exogenous monetary policy shocks with δt in carefullyidentified structural VAR models (see, e.g., Bernanke & Blinder, 1992; Blanchard & Quah, 1989; Christiano, Eichenbaum, & Evans,1999). However, even in the most sophisticated model, it is impossible to proxy for all information about future outputmovements that policymakers have had. Suppose that the true relationships between ∆Y and ∆m are represented in the system ofEqs. (1) and (2). Yet, the PBC's expectations about the future output movements are not observed. Thus, the vector E is left out ofthe system:ΔYt ¼ γ0 þ γ1Δmt þ γ2Zt þ ϑt; ð4ÞΔmt ¼ θ0 þ θ1ΔYt þ θ2Zt þ φt: ð5Þ2
4
6
8 5
10
15
20
25
98 99 00 01 02 03 04 05 06 07 08 09 10 11
Benchmark lending rate (LHS)
Discount rate (LHS)
Required reserve ratio (RHS)
Fig. 1. Three selected monetary policy instruments in China: the required reserve ratio, the benchmark lending rate and the discount rate (in %), 1998–2011a,b.
Notes: a. Starting from September 25, 2008, the reserve ratio for the small- and medium-sized financial institutions is set 1–2 percentage points lower than that
for the big financial institutions. In the figure, the ratio for the big financial institutions is reported. b. The discount rate is an interest rate at which the PBC lends to
commercial banks with a maturity of 20 days; the benchmark lending rate (with a maturity of 1 year) is an interest rates set by the PBC as a benchmark for banks
to follow. Gray shaded areas are marked as contractionary episodes, which will be elaborated in the following section.
Source: The minimum required reserve ratios are author's compilation based on the “Chronology” of various issues of the PBC's Annual Report. The interest rates
are from IMF International Financial Statistics.
R. Sun / China Economic Review 26 (2013) 56–74 61
The error terms in Eqs. (4) and (5) are ϑt = α3Et + !t and φt = β3Et + δt respectively. They are correlated. Clearly, the proxy
for exogenous monetary policy shocks with the error term φt will lead to biased estimates of monetary policy on output.
Obviously, this omitted-variable bias can be narrowed to some limit by including more control variables. However, the
analysis of the PBC's operating procedure and history suggests that the PBC watches an enormous number of economic variables
when setting policy. Many original macroeconomic data are not publicly available in China, neither are the PBC's numerical
forecasts of future economic developments. Thus, left-out variables are a serious problem in the estimates of the PBC's monetary
policy effects with simple regressions.
Another concern is that the PBC's operating procedures have experienced huge changes over time. Thus, its monetary policy
reaction function cannot be well described with time-invariant Eq. (2). Rather, the coefficients in its reaction function should vary
from episode to episode.
đang được dịch, vui lòng đợi..
