RDP 2010-05: Direct Effects of Money on Aggregate Demand: Another Look at the Evidence 1. Introduction

Money can be said to have direct effects if changes in the stock of money influence output and prices outside of its impact through short-term nominal interest rates. In general, it is an important issue whether or not money has such direct effects, but the issue is most prominent when the nominal interest rate approaches its zero lower bound, as it has in Japan, and more recently, in the euro area, the United Kingdom and the United States. With little or no room left to lower policy rates, central banks in these economies have moved to purchase a variety of assets with newly created money.[1] Figure 1 illustrates the recent increases in the real money base for the United Kingdom and the United States. Only if money has direct effects can this newly created money help to stimulate the economy.

Figure 1: Real Money Base

In most recent formulations of the canonical New Keynesian model, money has no direct effects; the equilibrium paths of output, inflation and the nominal interest rate are fully described and determined by three equations: one for aggregate demand, one for aggregate supply and a monetary policy rule for the short-term nominal interest rate. In the background of this equilibrium lies a money market with a central bank ready to supply as much money as needed for that market to clear at the interest rate called for by the policy rule. It is possible to augment this system of three equations and track monetary aggregates, but adding such an equation – whatever it may be – does not give money any direct effect over output or prices. With the nominal interest rate in the system, money is redundant. This is not to say, however, that money growth does not determine inflation in the long-run. This would generally be the case both in models where money has only indirect effects as well as in models where money has direct effects.[2] The difference lies in the mechanism through which a change in the stock of money causes a change in prices.

Monetarists would consider that this New Keynesian description of the transmission mechanism is, at best, incomplete. In their view – exemplified in Meltzer (2001) – movements in short-term nominal interest rates are insufficient to capture all the power of monetary policy actions. These actions are felt in many financial markets – beyond the interbank market for short-term debt – and generate wealth effects via various asset price movements that affect, through spending, both output and prices.[3]

The empirical evidence is mixed. Rudebusch and Svensson (1999, 2002) find little role for monetary aggregates in empirical aggregate demand specifications, while Favara and Giordani (2009) find that shocks to broad monetary aggregates have substantial and persistent effects on output. Hafer, Haslag and Jones (2007) find that money is not redundant, and Leeper and Roush (2003) find that whether money enters a model, and the way in which it does so, matters for inferences about policy impacts.

Nelson (2002) – complementing results in Koenig (1990) and Meltzer (2001) – shows that, after controlling for the short-term real interest rate, real base money growth is a significant determinant of total output in both the United Kingdom and the United States for the period 1960 to 1999. He then shows that an extended version of the New Keynesian model, in which the long-term nominal interest rate enters the money demand equation, can account for this finding. In particular, the extended model generates data which gives rise to positive and statistically significant coefficients of real money growth in regressions similar to those used on actual data.

As Nelson (2003) points out, however, this extension of the New Keynesian model does not introduce direct effects of money in the way that we describe them above. Money matters in Nelson's equations because it is an information variable, a proxy for other variables which are omitted. In this way, Nelson (2002) has redefined ‘direct effects’ in terms of the explanatory power for aggregate demand contained in the real money stock that is not captured in the short-term real interest rate. This definition is not particularly helpful, since money can be a function of many variables (and therefore contain information about them) but these many variables need not be a function of money. For base money expansions to stimulate the economy at the zero bound, money has to have structural direct effects and not informational ones.

The paper shows that the evidence both for and against structural direct effects, based on the statistical significance of the coefficients on real money growth in the regressions of Nelson (2002) and Rudebusch and Svensson (2002), is flawed. To do this, we follow Nelson (2002), take his extended model where money has no structural direct effects, generate data, and then estimate analogs of the empirical specifications. We also take another model – of Andrés, Lóepez-Salido and Nelson (2004) – where money has sizeable structural direct effects and do the same. We find that the model with no structural direct effects can give rise to informational direct effects and a model with structural direct effects can fail to do so. Our interpretation of the estimates is that the empirical specifications, by excluding a set of variables, introduce an omitted variables bias in the estimated coefficients on real money growth. These biases – although they could be thought to stand for informational effects – blur the structural relation between money and the rest of the economy. Although the regressions may at times reveal the existence of informational effects, they fail to uncover structural direct effects of money.

The rest of the paper is structured as follows. Section 2 updates and extends estimates of the specifications used by Nelson (2002) for the United States, the United Kingdom, Australia and Japan. Section 3 presents the extended model of Nelson (2002), which is used in Section 4 to conduct a Monte Carlo analysis. There, we generate data from the artificial economy, estimate empirical analogs of the specifications used in Section 2, and then decompose the bias of the real money growth coefficient. Section 5 concludes.

Footnotes

See Bernanke (2009), Borio and Disyatat (2009) and Dale (2010). [1]

We say generally because in the models of Krugman (1998) and Svensson (1999), households become willing to hoard any additional money that authorities choose to supply after the nominal interest rate reaches its lower bound. In this situation, there is no well-defined equilibrium level of real money balances and policy-makers lose control of the price level. The relation which links the growth rate of the money supply to the growth rate of prices can be thought to break down in a liquidity trap. [2]

See also Friedman (1956) and Brunner and Meltzer (1993). [3]