The identification assumption of the asset purchases in the DSGE models is particu- larly strong. Estimated VAR using the exogenous restriction imposed by the DSGE models suggests no evidence of a positive effect of asset purchases. This explains that the small effects of asset purchases found by the DSGE models may be due to the usage of the historical data to identify LSAPs as a supply shock. Estimated VAR with further relaxed DSGE restrictions (using only sign restrictions implied by the DSGE models) shows that asset purchases could potentially have a large effect on economy, but the identification scheme adopted by this VAR prevents further sharpening of the bounds of the effects.
Constructing DSGE models that are capable of correctly identifying the macro effects of the unconventional monetary policy from macro data is critical not only to assess the effectiveness of the policy but also to guide future exit strategies. Without understanding the transmission mechanism of those unconventional monetary policy to the macro economy, it is impossible to forecast how a future reversal in asset purchases or a raise in policy rate can impact the economy, and thus advise when the Federal Reserve should exit and how fast the pace of the sales of the assets should be. Neither large-scale asset purchase nor near zero-interest-rate policy is a
new experience. Japan has experienced near zero-interest-rate policy since 199956
and has adopted the “quantitative easing” policy between March 2001 and March 2006. Japan’s experience can provide economists valuable lessons to identify the transmission mechanism of those unconventional monetary policy. This points to my future research: using Japanese data to better identify the effect of asset purchases and better estimate the Markov-switching probability of the policy regime in the DSGE model. Japan also has experienced the exit of ZIRP twice, the abolishment of the quantitative easing, and the removal of the excess bank reserves. Like Japan before, the Federal Reserve now is running up an enormous balance sheet57and facing
the uncertainty surrounding an exit strategy. Possibly, Japan before and the United States now have something critical in common. Japan’s lesson will help forecast the evolution of the the Federal Reserve’s balance sheet going forward and the economic outlook.
Technically, the next step is to apply the perturbation method for Markov-switching models proposed by Foerster, Rubio-Ram´ırez, Waggoner, and Zha (2012). This method begins from the first principles and allows higher order approximation which may be important when taking into account the risk of the long-term bonds.
56Japan has experienced three ZIRP episodes: 1999Q2-2000Q2, 2001Q1-2006Q1, and 2010Q4-
present.
Figure 3.1: VAR identified by the exogenous restriction. The red lines show the mean of predicted path of the macro variables without shocks and under no intervention generated by the estimated VAR model using the DSGE exogenous restriction iden- tification. The blue lines show the mean of the predicted path of the macro variables under the LSAPs II generated by the same VAR model. The green lines show the mean of the predicted path ofthe macro variables under the ZIRP for four quarters generated by the same VAR model.
Figure 3.2: VAR identified by the exogenous restriction: effects of LSAPs. The red lines show the mean and the 90% Bayesian credible intervals of predicted path of the macro variables without shocks and under no intervention generated by the estimated VAR model using the DSGE exogenous restriction identification. The blue lines show the mean and the 90% Bayesian credible intervals of the predicted path of the macro variables under the LSAPs II generated by the same VAR model.
Figure 3.3: VAR identified by the exogenous restriction: effects of ZIRP.The red lines show the mean and the 90% Bayesian credible intervals of predicted path of the macro variables without shocks and under no intervention generated by the estimated VAR model using the DSGE exogenous restriction identification. The green lines show the mean and the 90% Bayesian credible intervals of the predicted path of the macro variables under the policy of keep interest rates at the 2010Q2 level (0.048%) for 4 quarters generated by the same VAR model.
Figure 3.4: VAR identified by exogenous restrictions. The red lines show the mean of predicted paths of the macro variables without shocks and under no intervention generated by the estimated VAR model using the DSGE exogenous restriction identi- fication. The blue lines show the mean of the predicted paths of the macro variables under the LSAPs II generated by the same VAR model. The green lines show the mean of the predicted path of the macro variables under the ZIRP for four quarters generated by the same VAR model. The grey lines are the predictive paths under the combination of the LSAPs and the ZIRP.
Figure 3.5: VAR identified by exogenous restrictions: combination of LSAPs and ZIRP.The red lines show the mean and 90% Bayesian credible intervals of predicted paths of the macro variables without shocks and under no intervention generated by the estimated VAR model using the DSGE exogenous restriction identification. The magenta lines are the mean and 90% Bayesian credible intervals of the predictive paths under the combination of the LSAPs and the ZIRP.
Figure 3.6: VAR identified by sign restrictions. The red line shows the mean of predicted path of macro variables without shocks and under no intervention generated by the estimated VAR model using the sign restriction identification. The blue line shows the mean of the predicted path of macro variables under the LSAPs II generated by the same VAR model. The green line shows the mean of the predicted path of macro variables under the ZIRP for four quarters generated by the same VAR model.
Figure 3.7: VAR identified by sign restrictions with identified set: effects of LSAPs.
The red lines show the mean and the identified set of predicted path of macro vari- ables without shocks and under no intervention generated by the estimated VAR model using the sign restriction identification. The blue lines show the mean and the identified set of the predicted path of macro variables under the LSAPs II generated by the same VAR model.
Figure 3.8: VAR identified by sign restrictions with identified set: effects of ZIRP.The red lines show the mean and the identified set of predicted path of macro variables without shocks and under no intervention generated by the estimated VAR model using the sign restriction identification. The green lines show the mean and identified set of the predicted path of macro variables under the LSAPs II generated by the same VAR model.
Figure 3.9: VAR identified by sign restrictions. The red lines show the mean of pre- dicted paths of macro variables without shocks and under no intervention generated by the estimated VAR model using the sign restriction identification. The blue lines show the mean of the predicted paths of the macro variables under the LSAPs II generated by the same VAR model. The green lines show the mean of the predicted paths of the macro variables under the ZIRP for four quarters generated by the same VAR model. The grey lines are the predictive paths under the combination of the LSAPs and the ZIRP.
Figure 3.10: Summary of effects of LSAPs and ZIRP in DSGE models and VAR mod- els. The squares stand for mean effects and the circles reflect the uncertainty. Green represents bonds-in-utility model, blue represents the results reported by chapter 1, pink represents the VAR with exogenous restrictions, and red represents the VAR with sign restrictions