I. Introduction

“The seventies were indeed special.” − Alan Blinder (1982)

It is widely documented that during the late 1960s and throughout the 1970s, inflation was high, volatile, and persistent, and a few recessions hit the U.S. economy.² This historical record, which is known as the “Great Inflation” episode, was followed by a period, beginning in the early 1980s, where the level, variance, and the persistence of the inflation rate, and the volatility of output decreased significantly. The latter phenomenon was labeled in the literature as the period of the “Great Moderation.” Some of the theories put forward to explain this historical record rely on “mistakes” of the Federal Reserve System (Fed) during the “Great Inflation” period.³ Most of these theories, however, have difficulties explaining why this record has changed since the early 1980s. At a deeper level, theories relying upon a change in the Fed’s behavior to explain the change in macroeconomic performance fail to explain why Fed behavior itself has changed.

In this paper, we outline a framework that can help explain the Great Inflation without relying on policy mistakes while, at the same time explaining why both macroeconomic performance and the Fed’s behavior have changed. The central ingredient in our analysis is the dramatic change in financial markets that took place around 1980, leading to more widespread participation in asset markets. We put together institutional evidence from a variety of sources showing that financial constraints were especially binding in the 1970s and that deregulation and financial innovation led to more widespread participation since the early 1980s. In contrast, we present a standard business-cycle model with limited asset-market participation that predicts that if asset-market participation is low, aggregate demand, contrary to conventional wisdom, is positively related to real interest rates. We show that in our theoretical model, this finding implies that Fed policy in the pre-1980 years was consistent with both equilibrium determinacy and welfare maximization. Furthermore, we argue that the change in the policy rule from passive pre-1980 to active post-1980 could have been an endogenous response to the change in market participation, required by optimal policy.

We use Bayesian estimation techniques to estimate our model on U.S. data on two samples, corresponding to the pre- (Paul) Volcker and the Volcker-(Alan) Greenspan periods. Our results indicate that there have been major changes in the U.S. economy between the two subsamples, pertaining to both the structure of the economy (deep parameters) and the stochastic environment (shock processes). Most notably, (i) the share of agents participating in asset markets has changed from a lower to a higher value, generating a change in the sign of the IS-curve slope from positive (contrary to standard theory) to negative (as predicted by standard theory); and (ii) the response of monetary policy changed from “passive” to “active.” Owing to these changes, the equilibrium was determinate throughout the whole period; moreover, monetary policy conduct was consistent with minimizing overall macroeconomic variability, as required by welfare maximization.

To our knowledge, the findings that the IS curve’s slope changed sign and, moreover, that this change came from a change in asset market participation are entirely novel and have striking implications for interpreting the Great Inflation and reassessing the Fed’s policy.⁴ Since our results show that shock processes have also changed, we run counterfactual experiments to study the relative importance of the “structure” versus “shocks” explanations of the changes in outcomes between the two subsamples. We find that although most of the changes can be accounted for by changes in the structure, changes in shock processes are also needed to explain some key facts (namely, the fall in the volatility of output). Finally, we show that the dynamic effects and propagation of fundamental shocks, in particular of “cost-push” and “aggregate demand” shocks, generated by our model are in line with stylized facts, and the bulk of the change in this transmission across the two samples is accounted for by the changes in asset-market participation and monetary policy conduct.

Our approach is most closely related to the large literature investigating the link between monetary policy and macroeconomic performance, with a particular focus on the Great Inflation and U.S. monetary policy in the 1970s. Some recent prominent contributions in this vein include Clarida, Galí, and Gertler (referred to as CGG) (2000), Taylor (1999), Lubik and Schorfheide (referred to as LS) (2004) and Ireland (2004). These studies estimate policy rules relating the policy instrument (a short term nominal interest rate) to macroeconomic variables such as expected inflation and output gap. All the cited papers identified a change in monetary policymaking with the coming to office of Paul Volcker as chairman of the Fed. Specifically, monetary policy was accommodative (“passive”) in the pre-Volcker years, increasing nominal interest rates less than one-to-one when expected inflation increased. In contrast, Fed policy was more restrictive (“active”) during the Volcker and Greenspan tenures. Since macroeconomic performance also changed, explaining the observed structural break by the change in the conduct of monetary policy became the norm in the profession.

The previously-mentioned studies argue that policy before Volcker was “badly” conducted along one or several dimensions, which led to worse macroeconomic performance as compared with the Volcker-Greenspan era. To make this point, estimated policy rules are embedded into calibrated general equilibrium models to study the dynamics and variability of macroeconomic variables. These theoretical predictions are then compared with stylized facts. CGG (2000) were the first to argue that the passive policy rule in the pre-Volcker sample led to equilibrium indeterminacy and left room for sunspot fluctuations, which instead led to a higher level and variability of inflation, and overall macroeconomic instability. This approach has four obvious difficulties, however, in explaining the Great Inflation: (i) sunspot shocks increase both inflation and output (and the output gap), something not seen in the data; (ii) in the theoretical model, the effects of fundamental shocks cannot be studied when equilibrium is indeterminate; (iii) the dynamics of the whole economy are entirely dependent upon the stochastic properties, the location, and the origin of the sunspot shock, all of which impossible to quantify in practice; and (iv) it is not clear why the Fed would have followed a policy that was so clearly suboptimal, given the model?⁵

The plan of our paper is as follows. In Section II, we outline the theoretical framework consisting of a standard “new-synthesis” model augmented for limited asset market participation and derive analytically its main theoretical implications. Section III provides empirical evidence of our hypothesis. First, we present institutional evidence on the change in U. S. asset market participation in the early 1980s. Then, we estimate a dynamic stochastic general equilibrium (DSGE) model with limited-asset market participation using Bayesian methods. In Section IV, we show that the estimated model is able to reproduce some of the stylized facts of the U.S. economy, most notably the changes in macroeconomic outcomes; we assess quantitatively the relative merits of explanations of these changes based on changes of the structure, on the one hand, and of the stochastic environment, on the other hand. Section V contains concluding remarks.

II. Limited Asset Market Participation and Monetary Policy: Some Theory.

In this section, we briefly outline a theory that allows the analysis of monetary policy under limited asset market participation while treating the degree of asset market participation as a parameter that can be exogenously influenced by policy. The framework is a modification of the by now standard dynamic sticky price cashless general equilibrium model, similar to the workhorse model in for example, Woodford (2003) or CGG (1999). The modification is that we allow for limited asset market participation, or “segmented asset markets”: part of the agents (asset holders) trade in complete asset markets including a market for shares in firms, while the other agents (non-asset holders) do not trade any assets and hence receive only a wage income. The share of non-asset holders, say λ, is exogenous, as in for example, Alvarez, Lucas, and Weber (2001).⁶ These agents will fail to smooth consumption as in Mankiw (2000) or Galí, López-Salido, and Valles (referred to as GLV) (2003b), where this comes from the failure to accumulate physical capital.

The model outlined here is related to the framework in GLV (2003b) and Bilbiie (2003). In contrast to GLV (2003b), however, we derive a model that abstracts from capital accumulation and focuses on a different set of questions namely, how the presence of non-asset holders alters the slope of the aggregate Euler equation (IS curve), determinacy properties of interest rate rules, optimal monetary policy, and the response of the model economy to various fundamental shocks. ⁷ In contrast to Bilbiie (2003), we use these theoretical insights to re-interpret the Great Inflation episode by estimating the model on U.S. data using Bayesian techniques and running counterfactual experiments. The chosen framework is well suited for our exercise for at least four reasons. First, it emphasizes the effect of non-asset holders on aggregate demand, which we wish to test empirically. Second, it derives analytically the “Inverted Taylor Principle” as a generically necessary condition for both equilibrium uniqueness and optimal policy when enough agents do not participate to asset markets. Third, it is directly comparable with and nests as a special case models such as CGG (2000) and LS (2004), which interpret the Great Inflation episode using estimated policy rules and comparing them to prescriptions dictated by theoretical models. Fourth, the absence of capital accumulation allows us to obtain analytical results and be transparent about the mechanism at work. Notably, optimal policy can be analyzed in a tractable way.

The exposition here is stripped down to the essential, ⁸ although the main results carry through to more general specifications of preferences and technology (see Bilbiie (2003) for details). A more general model that is estimated in the next section is described in detail in the Appendix.

Suppose aggregate expenditure consists of consumption only. There are two types of households: asset holders indexed by S, trading state-contingent assets and shares in firms and non-asset holders indexed by H, who do not participate in any of the asset markets and simply consume their current income. ⁹ The shares of these agents are 1 − λ and λ, respectively, and are assumed to be constant. Total consumption in log-linear deviations from steady state is given by c_t = λc_H,t + [1 − λ] c_S,t, where c_j,t is consumption of group j.¹⁰ Suppose furthermore for simplicity that labor supply of non-asset holders is inelastic n_H,t = 0, such that their consumption is equal to the real wage c_H,t = w_t and total labor supply is given by n_t = [1 − λ] n_S,t. Assume that asset holders” labor supply obeys a standard optimality condition φn_S,t = w_t − c_S,t, where φ is the inverse Frisch elasticity of labor supply for type S. Total consumption will hence be: $c_t={\lambda }{w}_t+[1-{\lambda }]c_{S,t}={\lambda }{\phi }_S{n}_{S,t}+c_{S,t}=\frac{{\lambda }}{1-{\lambda }}\phi n_t+c_{S,t}$. Finally, assume that the production function for final output in log-linear form is y_t = [1 + µ] n_t, where µ represents both the steady-state net mark-up and the degree of aggregate increasing returns to scale. Using this we obtain a version of the “planned expenditure” or “aggregate demand” equation from standard Keynesian models:¹¹

This equation links aggregate expenditure to current income and consumption of asset holders. Note that (1) is not a reduced-form relationship since c_t, y_t, c_S,t are all endogenous variables, which will be determined in general equilibrium. However, we can think of (1) as a schedule in the (y, c) space, for a given level of c_S,t. In that sense, we can say that aggregate demand (expenditure) depends positively on current income and negatively on the real interest rate. We can define the (partial) “marginal propensity to consume” out of current income as $\partial c/\partial y=\frac{{\lambda }}{1-{\lambda }}\frac{\phi }{1+\mu }>0$. This “marginal propensity to consume” is in fact a partial marginal propensity, that is keeping fixed consumption of asset holders c_S. In equilibrium, of course, all output is consumed. We will loosely refer to ∂c/∂y as “marginal propensity to consume” in the remainder. The negative impact of ex-ante real interest rates rr_t ≡ r_t − E_tπ_t+1 on aggregate demand comes from a standard Euler equation for consumption of asset-holders: c_S,t = E_tc_S,t+1− [r_t − E_tπ_t+1], where r is the nominal interest rate and the intertemporal elasticity of substitution in consumption is normalized to one without loss of generality.

The marginal propensity to consume depends positively on (i) the share of non-asset holders λ and (ii) the elasticity of labor supply φ. The impact of the former on the marginal propensity can be rationalized by the dependence of total consumption upon consumption of agents who do not hold assets and are unable to smooth consumption, and are therefore sensitive to changes in current income. With regards to the labor supply elasticity, note that inelastic labor supply implies small variations in hours (and output) and is associated with large variation in real wage and hence consumption of non-asset holders. Consequently, the aggregate propensity to consume depends finally on income distribution, which changes as aggregate income and the wage rate change; this gives the model a distinctly Keynesian flavor. Together with the condition that consumption equal output c_t = y_t, equation (1) leads to a “Keynesian cross”-type diagram and the standard IS equation in case ∂c/∂y < 1 (see the thick black line labelled “K” in Figure 1).

Keynesian and Non-Keynesian Crosses.

Citation: IMF Working Papers 2006, 200; 10.5089/9781451864601.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Keynesian and Non-Keynesian Crosses.

Citation: IMF Working Papers 2006, 200; 10.5089/9781451864601.001.A001

• Download Figure
• Download figure as PowerPoint slide

Keynesian and Non-Keynesian Crosses.

Citation: IMF Working Papers 2006, 200; 10.5089/9781451864601.001.A001

• Download Figure
• Download figure as PowerPoint slide

However, note that the marginal propensity to consume out of current income (output) ∂c/∂y can become greater than one. This case, which we label “non-Keynesian”, occurs when enough agents consume their wage income w_t (λ high) and/or wage is sensitive enough to real income y_t (φ high), more precisely when:

Aggregate marginal propensity to consume (again, for a given c_S) can be larger than one since non-asset holders consume all their current income given by the real wage, which is instead related to total output (income) more than one-to-one (and the more so, the more inelastic is labor supply).

We plot (1) in this case along with the c = y schedule in the “Non-Keynesian cross” in Figure 1, where an increase in the real interest rate moves the (1) schedule rightward (by intertemporal substitution) leading to higher consumption and output.

An immediate implication of (2) is that the slope of the aggregate IS curve changes sign. Consumption of asset holders is related to total output, combining (1) with c_t = y_t, by:

Note that δ becomes negative when ∂c/∂y >1, that is precisely when (2) holds. Consumption of asset holders can be negatively related to total output since an increase in demand can only be satisfied by movements of (as opposed to movements along) the labor supply schedule when enough people hold no assets and labor supply is inelastic enough. But the necessary rightward shift of labor supply can only come from a negative income effect on consumption of asset holders. This negative income effect is ensured in general equilibrium by a potential fall in dividend income. Note that asset holders have in their portfolio (1 − λ)⁻¹ shares: if total profits fell by one unit, dividend income of one asset holder would fall by (1 − λ)⁻¹ > 1 units.¹² The potential decrease in profits is a natural result of inelastic labor supply, since the increase in marginal cost (real wage) would more than outweigh the increase in sales (hours). Therefore, the mechanism relies on consumption of asset holders being sensitive to unforeseen changes in the money-value of their assets¹³.

Substituting (3) into the Euler equation we obtain the aggregate IS curve:

Direct inspection of (4) suggests the impact that limited asset market participation has on the dynamics of a standard business cycle model through modifying the elasticity of aggregate demand to real interest rates δ⁻¹ in a non-linear way. For high enough participation rates λ < λ* (where the latter is given by (2)) we are in a “Keynesian” region, whereby real interest rates restrain aggregate demand. As λ increases towards λ*, the sensitivity to interest rates increases in absolute value, making policy more effective in containing demand. However, once λ is above the threshold λ*, we move to the “non-Keynesian” region where increases in real interest rates become expansionary (see also Figure 1). As λ tends to its upper bound of 1, δ⁻¹ decreases towards zero, that is policy is ineffective when nobody holds assets. We will call “non-Keynesian” an economy in which participation in asset markets is limited enough such that δ < 0. Finally, note that the only way for δ to be independent of λ is for φ to be zero, that is labor supply of asset holders be infinitely elastic. In this case, consumption of all agents is independent of wealth, making the heterogeneity introduced in this paper irrelevant.

III. Empirical Evidence

In this section, we provide empirical evidence for the discussed hypothesis. First, we present institutional evidence for the change in U.S. asset market participation in the early 1980s. Further, we estimate a version of our model with a richer dynamic structure on U.S. data by utilizing Bayesian structural estimation techniques,²² distinguishing between the pre-Volcker and Volcker-Greenspan samples. We show that, consistently with the institutional evidence, the fraction of agents participating in asset markets has changed between the two periods, causing a significant change in the sign of the sensitivity of aggregate demand to interest rates. As we argued before, we believe that this is an important and so far neglected part of the “Great Inflation” story. Consistently with the results of other papers, we also find that the response of monetary policy changed from passive to active between the two samples, several other structural changes detailed below took place and, importantly, that the distribution of the shocks has changed. Finally, we investigate whether the changes in “structure” (deep parameters) or “shocks” (stochastic environment) have been paramount in driving the changes in U.S. macroeconomic outcomes.

B. Bayesian Estimation of Structural Model

In this section we fit a general version of our model to quarterly U.S. data on real GDP, the annualized quarterly CPI inflation, and the quarterly average of the overnight federal funds rate, expressed in units of percent per year.²⁶ To make our results comparable with previous studies we divide the data set into two subsamples. The pre-Volcker sample²⁷ contains quarterly data from 1965:IV to 1979:III, while the Volcker-Greenspan sample reaches from 1983:I to 2002:I. In line with the related literature, we exclude the “Volcker disinflation” period from the sample. To capture the underlying dynamics in macroeconomic variables, we extend our baseline model to allow for less restrictive endogenous persistence. In particular, we introduce, in line with the empirical literature, habit persistence in consumption and price-indexation into the model, while monetary policy formulation is characterized by a more general Taylor rule with interest rate smoothing. ²⁸ The model is presented in detail in the Appendix.

The canonical (or reduced-form) representation of the estimated model is basically indistinguishable from the extended version of the model estimated by LS (2004). However, the introduction of non-asset holders changes the underlying elasticities and, in contrast to the standard literature, allows for a positive IS-curve slope. Namely, the corresponding IS-curve with habit persistence and limited asset market participation has the following form:

$\begin{array}{ccc}{y}_t=\frac{{{\Gamma }}_1}{{{\Gamma }}_1+{{\Gamma }}_2}{E}_t{y}_{t+1}+\frac{{{\Gamma }}_2}{{{\Gamma }}_1+{{\Gamma }}_2}{y}_{t-1}-\frac{1-\gamma }{{{\Gamma }}_1+{{\Gamma }}_2}(r_t-E_t{\pi }_{t+1})+g_t,& & (11)\end{array}$

where

${{\Gamma }}_1=1-\frac{{\lambda }}{1-{\lambda }}\frac{\phi }{1+\mu }\left[1+\frac{\gamma \mu }{1+\phi (1-\gamma )}\right];{{\Gamma }}_2=\gamma \left[1-\frac{{\lambda }}{1-{\lambda }}\frac{\phi }{1+\phi (1-\gamma )}\right].$

Notice that γ captures the degree of habit persistence in consumption, and influences the threshold level λ of beyond, which the slope of the IS curve changes sign.²⁹ In order to perform the estimation exercise, we enrich the stochastic structure of the model following LS (2004) and augment the IS-curve with an AR(1) “aggregate demand” shock $g_t={\rho }^g{g}_{t-1}+{\epsilon }_t^g$. Similarly, we add to the New-Keynesian Phillips curve with limited asset market participation and price indexation an AR(1) “supply” shock $z_t={\rho }^z{z}_{t-1}+{\epsilon }_t^z$. The Phillips curve is:

$\begin{array}{ccc}{\pi }_t=\frac{\beta }{1+\beta \omega }{E}_t{\pi }_{t+1}+\frac{\omega }{1+\beta \omega }{\pi }_{t-1}+\psi \left(1-{\gamma }+\frac{\phi }{1+\mu }\right)y_t-\psi \frac{{\gamma }}{1-{\gamma }}{y}_{t-1}+z_t,& & (12)\end{array}$

where ω is the degree of price indexation, and ψ ≡ (1 − θ)(1 − θβ)/θ. Importantly, notice that ε_g,t is a non-structural reduced-form shock, and represents a convolution of shocks to technology, government spending, preferences, and other shocks apart from cost-push. As discussed for example in Woodford (2003), all these shocks will also have an immediate, direct impact on the Phillips curve. On the contrary, there exists shocks that can potentially move the Phillips curve without having any impact on the IS curve (the so-called “cost-push” shocks). We model this, following LS (2004) by assuming that all potential shocks that move the IS curve, summarized by ε_g,t also influence influence the Phillips curve; that is, in the estimation we allow for a non-zero correlation ρ^gz between the innovations ε_g,t, ε_z,t. Correspondingly, in what follows the standard deviations of the reduced-form innovations are defined as ${\sigma }^{{\epsilon }_g},{\sigma }^{{\epsilon }_z}$, while the standard deviations of the structural shocks, say e_g and e_z can be found by standard Choleski-decomposition algebra: ${\sigma }^{e_g}={\sigma }^{{\epsilon }_g}$ and ${\sigma }^{e_z}={\sigma }^{{ϵ}_z}\sqrt{1-{\left({\rho }^{gz}\right)}^2}$. Note that the “supply” shock to the Phillips curve will be a linear combination of the “IS” shock e_g and the cost-push shock e_z:

The model is closed by a general version of a Taylor rule, incorporating interest rate smoothing:

$\begin{array}{ccc}{r}_t={\varphi }_r{r}_{t-1}+(1-{\varphi }_r)({\varphi }_{\pi }{E}_t{\pi }_{t+1}+{\varphi }_x{y}_t)+{\epsilon }_t^r,& & (13)\end{array}$

where the monetary policy shock ${\epsilon }_t^r$ is white noise.

The theoretical model is a system of log-linearized equations. To connect it to the available dataset, we will undertake the following transformations. First, we remove the smooth trend from the real U.S. GDP series using the Hodrick-Prescott (HP) filter. Then, we connect the vector of endogenous variables s_t = [y_t, π_t, r_t] to the vector of observable variables o_t via the following measurement equations:

$o_t=\left[\begin{array}{c}0\\ {\pi }^{*}\\ r^{*}+{\pi }^{*}\end{array}\right]+\left[\begin{array}{lll}1& 0& 0\\ 0& 4& 0\\ 0& 0& 4\end{array}\right]s_t,$

where π* and r* are annualized steady-state inflation and real interest rate in percentage terms. The measurement equation together with the structural equations and the shock processes form the state space representation of the observables o_t. The parameter vector has the form: $\Phi =[r^{*},{\pi }^{*},\lambda ,\phi ,\mu ,\gamma ,\omega ,\beta ,{\varphi }_{\pi },{\varphi }_y,{\varphi }_r,{\rho }^g,{\rho }^z,{\sigma }^{{\epsilon }_g},{\sigma }^{{\epsilon }_z},{\sigma }^{{\epsilon }_r},{\rho }^{gz}]$. We use the Kalman-Filter to evaluate the corresponding likelihood function L(y^T|Φ) of the state space representation of the model. For the estimation, we adopt a Bayesian approach and combine the likelihood of the model with a prior density p(Ф). By neglecting any constants the posterior density function has the following form:

$\begin{array}{ccc}p(\Phi |Y)=L(Y^T|\Phi )p(\Phi ).& & (14)\end{array}$

There is no closed form solution for the posterior p(Φ|Y), so we calculate the posterior distribution of the parameters using the Metropolis-Hastings sampling algorithm. Before discussing the results, we present the choice of our priors.

Prior distributions

A number of parameters were kept fixed from the start. First, notice that by definition, the discount factor β equals (1 + r*/100)^−1/4. We also set the steady state mark up µ to 0.2 and the degree of nominal rigidity θ to 0.75. Both values are frequently chosen in calibration exercises. Table 1 provides details about the set of prior distributions for the remaining parameters.

Table 1.

Prior Distributions for DSGE Model Parameters

Notes: The prior for ρ^gz is truncated to ensure that the correlation lies between -1 and 1.

Table 1.

Prior Distributions for DSGE Model Parameters

		Pre-Volcker		Volcker-Greenspan
Name	Density	Mean	Std. Deviations	Mean	Std. Deviations
λ	Beta	0.35	0.10	0.30	0.10
φ	Gamma	3.00	0.50	3.00	0.50
γ	Beta	0.50	0.20	0.50	0.20
ω	Beta	0.50	0.20	0.50	0.20
ϕπ	Gamma	0.50	0.50	2.00	0.50
ϕy	Gamma	0.25	0.15	0.25	0.15
ϕr	Beta	0.25	0.25	0.25	0.25
π*	Gamma	4.00	2.00	4.00	2.00
r*	Gamma	2.00	1.00	2.00	1.00
ρg	Beta	0.70	0.10	0.70	0.10
ρz	Beta	0.70	0.10	0.70	0.10
${\sigma }^{{ϵ}_g}$	Inverse Gamma	0.38	0.20	0.38	0.20
${\sigma }^{{ϵ}_z}$	Inverse Gamma	1.00	0.52	1.00	0.52
${\sigma }^{{ϵ}_r}$	Inverse Gamma	0.31	0.16	0.31	0.16
ρgz	Truncated Normal	0.00	0.10	0.00	0.10

Notes: The prior for ρ^gz is truncated to ensure that the correlation lies between -1 and 1.

View Table

Table 1.

Prior Distributions for DSGE Model Parameters

		Pre-Volcker		Volcker-Greenspan
Name	Density	Mean	Std. Deviations	Mean	Std. Deviations
λ	Beta	0.35	0.10	0.30	0.10
φ	Gamma	3.00	0.50	3.00	0.50
γ	Beta	0.50	0.20	0.50	0.20
ω	Beta	0.50	0.20	0.50	0.20
ϕπ	Gamma	0.50	0.50	2.00	0.50
ϕy	Gamma	0.25	0.15	0.25	0.15
ϕr	Beta	0.25	0.25	0.25	0.25
π*	Gamma	4.00	2.00	4.00	2.00
r*	Gamma	2.00	1.00	2.00	1.00
ρg	Beta	0.70	0.10	0.70	0.10
ρz	Beta	0.70	0.10	0.70	0.10
${\sigma }^{{ϵ}_g}$	Inverse Gamma	0.38	0.20	0.38	0.20
${\sigma }^{{ϵ}_z}$	Inverse Gamma	1.00	0.52	1.00	0.52
${\sigma }^{{ϵ}_r}$	Inverse Gamma	0.31	0.16	0.31	0.16
ρgz	Truncated Normal	0.00	0.10	0.00	0.10

Notes: The prior for ρ^gz is truncated to ensure that the correlation lies between -1 and 1.

In most of the cases, we choose the same priors for the parameters for both, the pre-Volcker and Volcker-Greenspan period. We deviated from this strategy only for the parameters capturing the share of liquidity constraint agents λ and the coefficient governing the interest rate response to changes in expected inflation ϕ_π. There is overwhelming empirical evidence that the interest rate response to inflation in the Fed’s monetary rule was below one for the pre-Volcker period. ³⁰ Accordingly, we chose the corresponding prior to follow a gamma distribution with mean 0.5 for the pre-Volcker period, while we set the prior mean to be equal to 2 for the Volcker-Greenspan era. By setting the standard deviation equal to 0.5, we chose a wide confidence interval for this parameter. Furthermore, in line with the institutional evidence presented in the previous section, we set the mean of the distribution describing the share of liquidity constraint agents to a slightly higher value in the pre-Volcker era than in the Volcker-Greenspan period. In particular, we chose as a prior a beta distribution with mean 0.35 for the pre-Volcker period, while the mean of the prior-distribution is assumed to be slightly lower at 0.30 following Volcker’s inauguration. We believe that the choice of different priors is justified not only because, as discussed in the theoretical part of the model, it is required to ensure that the estimated model is determinate, but also given the institutional evidence provided in the previous subsection.

For the inverse of the elasticity of labor supply φ, we have chosen a gamma distribution with mean 3.00 and standard deviation 0.50 in both periods. The confidence interval for φ entails the values generally chosen in calibration exercises and consistent with microeconomic evidence (see for example Domeij and Floden, 2004). The priors for the habit persistence and the price indexation coefficients are both assumed to have a beta distribution with mean 0.5 and a standard deviation of 0.2. The prior for the output response coefficient in the monetary rule is assumed to have a gamma distribution with mean 0.25 and standard deviation 0.10. Similarly, the prior for the interest rate smoothing coefficient follows a beta distribution with mean 0.25 and standard deviation 0.25.

Estimation results

We present the estimation results in Table 2. Our hypothesis that the slope of the IS-curve changed sign in the 1980s is supported by the outcome. The posterior in both cases indicates that our prior for the share of non-asset holders was rather too low (in case of the pre-Volcker) or too high (in case of the Volcker-Greenspan era). In particular, the posterior mean of the share of non-asset holders λ falls from 0.44 in the pre-Volcker period to 0.24 in the Volcker-Greenspan era (notice that the 90 percent intervals for the two estimates do not overlap at all)³¹. Since we can assume that the data updated our priors into the “right” direction, the estimated results are in favor of the hypothesis that the share of non-asset holders, interpreted as a proxy for the degree of U.S. financial regulation, decreased in the 1980s. This change drives the change in the sign of the slope of the IS curve³², since the estimates of the posterior mean of the inverse of labor supply elasticity of asset holders φ are virtually unchanged: 2.98 in the pre-Volcker era and 2.86 in the Volcker-Greenspan period. Also, the estimated confidence interval entails in both cases the corresponding estimated posterior mean of the other period. Moreover, the estimates for the posterior mean of the degree of habit formation - another parameter that enters the slope of the IS curve - are also very close (0.43 and 0.36, respectively).

Table 2.

Bayesian Estimation Results

Notes: The slope of the IS-curve is calculated by using the formula described in equation (2.1) and the posterior mean of the estimated structural parameters.

Table 2.

Bayesian Estimation Results

	Pre-Volcker		Volcker-Greenspan
Parameter	Mean	90 percent interval	Mean	90 percent interval
λ	0.44	[0.38 – 0.51]	0.24	[0.17 – 0.31]
φ	2.98	[2.27 – 3.63]	2.62	[1.90 – 3.36]
γ	0.43	[0.24 – 0.64]	0.36	[0.27 – 0.45]
ω	0.54	[0.34 – 0.74]	0.07	[0.01 – 0.13]
IS-slope	1.25		-0.65
ϕπ	0.52	[0.42 – 0.62]	2.39	[2.09 – 2.70]
ϕy	0.10	[0.05 – 0.15]	0.07	[0.01 – 0.13]
ϕr	0.01	[0.00 – 0.02]	0.66	[0.57 – 0.75]
π*	4.54	[3.11 – 6.02]	3.26	[2.48 – 4.10]
r*	1.21	[0.43 – 2.03]	2.73	[1.98 – 3.47]
ρg	0.71	[0.60 – 0.83]	0.91	[0.87 – 0.96]
ρz	0.76	[0.67 – 0.87]	0.68	[0.55 – 0.80]
${\sigma }^{{ϵ}_g}$	0.29	[0.18 – 0.40]	0.16	[0.11 – 0.22]
${\sigma }^{{ϵ}_z}$	0.72	[0.57 – 0.87]	0.83	[0.67 – 0.98]
${\sigma }^{{ϵ}_r}$	0.34	[0.28 – 0.40]	0.15	[0.12 – 0.18]
ρgz	0.05	[(−0.11) – 0.21]	0.31	[0.15 – 0.45]

Notes: The slope of the IS-curve is calculated by using the formula described in equation (2.1) and the posterior mean of the estimated structural parameters.

View Table

Table 2.

Bayesian Estimation Results

	Pre-Volcker		Volcker-Greenspan
Parameter	Mean	90 percent interval	Mean	90 percent interval
λ	0.44	[0.38 – 0.51]	0.24	[0.17 – 0.31]
φ	2.98	[2.27 – 3.63]	2.62	[1.90 – 3.36]
γ	0.43	[0.24 – 0.64]	0.36	[0.27 – 0.45]
ω	0.54	[0.34 – 0.74]	0.07	[0.01 – 0.13]
IS-slope	1.25		-0.65
ϕπ	0.52	[0.42 – 0.62]	2.39	[2.09 – 2.70]
ϕy	0.10	[0.05 – 0.15]	0.07	[0.01 – 0.13]
ϕr	0.01	[0.00 – 0.02]	0.66	[0.57 – 0.75]
π*	4.54	[3.11 – 6.02]	3.26	[2.48 – 4.10]
r*	1.21	[0.43 – 2.03]	2.73	[1.98 – 3.47]
ρg	0.71	[0.60 – 0.83]	0.91	[0.87 – 0.96]
ρz	0.76	[0.67 – 0.87]	0.68	[0.55 – 0.80]
${\sigma }^{{ϵ}_g}$	0.29	[0.18 – 0.40]	0.16	[0.11 – 0.22]
${\sigma }^{{ϵ}_z}$	0.72	[0.57 – 0.87]	0.83	[0.67 – 0.98]
${\sigma }^{{ϵ}_r}$	0.34	[0.28 – 0.40]	0.15	[0.12 – 0.18]
ρgz	0.05	[(−0.11) – 0.21]	0.31	[0.15 – 0.45]

Notes: The slope of the IS-curve is calculated by using the formula described in equation (2.1) and the posterior mean of the estimated structural parameters.

Our results also confirm the results of inter alia CGG and LS by finding an inflation response below 1 in the interest rate rule in the pre-Volcker period. The estimated 90 percent interval goes from 0.42 to 0.62. The range for the response coefficient in the Volcker-Greenspan era reaches from 1.90 to 3.36. It should be emphasized, however, that while our results do indicate that the observational implications of monetary policymaking, as described by the Taylor rule, have changed, this does not necessarily imply that policymaking has changed at a deeper level. Indeed, in light of our theoretical results, the estimates indicate that monetary policy has been conducted with a view to minimize macroeconomic variability throughout the whole sample, since this strategy requires precisely a passive policy rule in the pre-Volcker period; in this sense, our results imply that there was no fundamental change in monetary policy, and are hence consistent with the findings of Sims and Zha (2006).

There are, however, several other factors worth discussing in further detail. The coefficient determining the degree of interest rate smoothing in the monetary policy rule differs significantly across the two periods. While the pre-Volcker era estimate indicates that there was almost no interest rate smoothing in the 1970s, the Federal Reserve had the tendency to adjust interest rates only gradually in response to changes in economic conditions in the Volcker-Greenspan era. If we believe that one important benefit of interest rate smoothing is to guide agents expectations about future interest rate developments, then the observed discrepancies can also be rationalized by the different degree of asset market participation in the two periods. A common argument (see, e.g., Woodford, 2003) is that the Fed can achieve its stabilization goals only insofar as its actions affect long-term interest rates. Hence, under rational expectations and full asset market participation, an effective response by the Fed requires that the private sector be able to believe that an increase in the short-term policy rate does also change the entire future path of interest rates. Obviously, this effect is more pronounced in an environment with high degree of asset market participation. As a result, it is not surprising that the degree of interest rate smoothing was significantly lower in the pre-Volcker era, where asset market participation was limited. The estimates of the degree of price-indexation do also differ significantly in the two periods: price-indexation was much more pronounced in the pre-Volcker period, which is in line with the stylized fact that inflation was highly persistent during this time.

Our results indicate that the two samples are also characterized by very different stochastic environments, which is consistent with the findings of a series of papers using different estimation techniques, such as Sargent, Williams, and Zha (2006) or Sims and Zha (2006), to quote the most recent examples. The persistence of the “demand” shocks is higher in the Volcker-Greenspan period, while the opposite is true for supply shocks. The correlation between supply and demand shocks is small in the pre-Volcker sample, but relatively large in the Volcker-Greenspan period, while the standard deviations of demand and monetary policy shocks have been falling. The standard deviation of “pure” cost push shocks ${\sigma }^{e_z}$ can be calculated from the standard deviation of the estimated, reduced-form supply shocks ${\sigma }^{{ϵ}_z}$ using the standard formula for Choleski decomposition ${\sigma }^{e_z}={\sigma }^{{ϵ}_z}\sqrt{1-{\left({\rho }^{gz}\right)}^2}$. This delivers 0.72 for the pre-Volcker and 0.78 for the Volcker-Greenspan periods respectively, suggesting that the standard deviation of cost-push shocks has been roughly stable.

All in all, our results indicate that there were changes in the U.S. economy in both structure and the distribution of shocks. Moreover, we identify some changes that were not identified before, most notably the change in asset market participation. Importantly, our results indicate that the structural changes were multidimensional, such that equilibrium determinacy was a feature of both the pre-Volcker and the Volcker-Greenspan samples. The crucial factors driving this result are the highly limited asset market participation combined with the passive monetary policy rule in the pre-Volcker sample. Otherwise put, if either policy were active or asset market participation were more widespread, equilibrium would have been indeterminate. As a result, the popular explanation (see, e.g., CGG, 2000 and LS, 2004) that sunspot shocks were the main source of macroeconomic volatility in the pre-Volcker era is not conducive with our hypothesis. ³³

Naturally, our framework implies that explanations for the Great Inflation, its conquest, and the difference in macroeconomic outcomes more generally, should be looked for elsewhere: either in the different distributions of fundamental shocks, or in the different economic structures in the two periods. In the former vein, many authors have argued (see for example Sargent, 2002) and papers quoted therein) that the two sub-samples were characterized by different stochastic environments (see also Cogley and Sargent, 2002). More specifically, Blinder (1982) argues that “the seventies were indeed special”, regarding the relative size of supply shocks (see also Stock and Watson, 2003). The latter interpretation suggests that changes in the structure of the economy are crucial in driving the change in outcomes. Our paper contributes to this “structure vs. shocks” debate by investigating the relative role of each of these hypotheses in the following section.

IV. A Change in Structure of Economy or in Distribution of Shocks?

In columns 1 and 3 of Table 3, we first present the stylized facts pertaining to U.S. output, inflation, and interest rate dynamics for the two chosen sub-samples. ³⁴ The results confirm the conventional wisdom that inflation was higher and more persistent and both inflation and output were much more volatile in the pre-Volcker era. In the next step, we analyze whether our model is able to reproduce some of these stylized facts. Therefore, in columns 2 and 4 of Table 3, we report the results of stochastic simulations using the estimated posterior mean of the moments of the shocks and the estimated posterior mean of the structural parameters for the pre-Volcker and Volcker-Greenspan periods, respectively. The results show, consistently with the data, that inflation was higher on average, more volatile and much more persistent in the earlier, “Great Inflation” period. The model also delivers the fall in the volatility of output that came to be known as the “Great Moderation”, as well as the fall in the volatility in the interest rates.

Table 3.

Stochastic Simulations

Notes: The “structure scenario” in column (5) presents the outcomes that would have occurred if the stochastic environment were the pre-Volcker one, but the structure of the economy were as in the Volcker-Greenspan sample. Column (6), which we call the “shocks scenario,” shows the moments that would have occurred if the structure of the economy were the pre-Volcker one, but the economy were subject to the same shocks as in the Volcker-Greenspan period.

Table 3.

Stochastic Simulations

	Pre-Volcker		Volcker-Greenspan		Scenario
	Data (1)	Model (2)	Data (3)	Model (4)	“Structure” (5)	“Shocks” (6)
Means
Output	0.01	0.00	0.01	0.00	0.00	0.00
Inflation	6.28	4.46	3.17	3.25	3.25	4.40
Interest Rate	6.56	5.71	6.21	5.98	5.99	5.67
Standard Deviations
Output	1.26	1.25	0.97	0.98	1.51	1.01
Inflation	3.09	4.66	1.54	2.27	3.21	5.96
Interest Rate	2.21	2.56	2.10	2.03	1.63	2.09
Persistence, AR(1)
Output	0.75	0.72	0.78	0.80	0.73	0.71
Inflation	0.89	0.88	0.46	0.52	0.50	0.94
Interest Rate	0.91	0.55	0.98	0.91	0.75	0.88

Notes: The “structure scenario” in column (5) presents the outcomes that would have occurred if the stochastic environment were the pre-Volcker one, but the structure of the economy were as in the Volcker-Greenspan sample. Column (6), which we call the “shocks scenario,” shows the moments that would have occurred if the structure of the economy were the pre-Volcker one, but the economy were subject to the same shocks as in the Volcker-Greenspan period.

View Table

Table 3.

Stochastic Simulations

	Pre-Volcker		Volcker-Greenspan		Scenario
	Data (1)	Model (2)	Data (3)	Model (4)	“Structure” (5)	“Shocks” (6)
Means
Output	0.01	0.00	0.01	0.00	0.00	0.00
Inflation	6.28	4.46	3.17	3.25	3.25	4.40
Interest Rate	6.56	5.71	6.21	5.98	5.99	5.67
Standard Deviations
Output	1.26	1.25	0.97	0.98	1.51	1.01
Inflation	3.09	4.66	1.54	2.27	3.21	5.96
Interest Rate	2.21	2.56	2.10	2.03	1.63	2.09
Persistence, AR(1)
Output	0.75	0.72	0.78	0.80	0.73	0.71
Inflation	0.89	0.88	0.46	0.52	0.50	0.94
Interest Rate	0.91	0.55	0.98	0.91	0.75	0.88

Notes: The “structure scenario” in column (5) presents the outcomes that would have occurred if the stochastic environment were the pre-Volcker one, but the structure of the economy were as in the Volcker-Greenspan sample. Column (6), which we call the “shocks scenario,” shows the moments that would have occurred if the structure of the economy were the pre-Volcker one, but the economy were subject to the same shocks as in the Volcker-Greenspan period.

In order to assess the relative importance of changes in the structure of the economy (as described by the deep parameters) versus those in the stochastic environment (captured by parameters pertaining to shock processes), we conduct two counterfactual simulations, reported in columns 5 and 6 of Table 3. Column 5, dubbed the “structure scenario,” shows the outcomes that would have occurred if the stochastic environment were the pre-Volcker one, but the structure of the economy were as in the Volcker-Greenspan sample. Column 6, which we call the “shocks scenario,” shows the moments that would have occurred if the structure of the economy (the deep parameters) were at their pre-Volcker values, but the economy were subject to the same shocks as in the Volcker-Greenspan period. The scenario that turns out to be closest to the actual Volcker-Greenspan outcomes would then indicate that the respective change (in shocks or structure, respectively) was relatively more important in explaining the change in outcomes. The results indicate that the “structure scenario” does better than the “shocks scenario”, as it delivers the fall in the mean and persistence of inflation, the increase in average interest rates, and for both, inflation and interest rates the fall in volatility. For all these moments the “shocks” scenario predicts movements in the opposite direction. However, the “shocks scenario” is crucial in explaining the “Great Moderation” (the fall in output volatility) and, to a lesser extent, the increase in interest rate persistence.³⁵ To summarize, both the structural change and the altering distribution of the shocks are important, albeit to different degrees, in order to explain the features of the Great Inflation and the change in macroeconomic dynamics in the Volcker-Greenspan era.

A. Asset Market Participation, Monetary Policy, and Propagation of Shocks

Our final experiments document the differences in the propagation of cost-push and “demand” shocks across the two samples, and the potential explanatory role of the changes in asset market participation and monetary policy. Figure 2 plots the impulse responses of output, inflation, and the nominal and real interest rate to a unit cost-push shock under three different scenarios.³⁶ The pre-Volcker and the Volcker-Greenspan scenarios use the estimated posterior mean of the parameters to calculate the dynamic behavior of the model economy. The responses of the model confirm both the conventional wisdom and what we view as a good test for a theory purported to explain dynamics in that period: higher -and more persistent- inflation, low real rates, and negative comovement of inflation and the output. Moreover, responses of output and inflation have the same sign under both scenarios, but in the pre-Volcker scenario the response of inflation is larger and the response of output more negative. Notice also that in the pre-Volcker period, the real interest rate is persistently negative since the policy rule is passive.

Impulse Response Functions Following a Cost-Push Shock

Citation: IMF Working Papers 2006, 200; 10.5089/9781451864601.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Impulse Response Functions Following a Cost-Push Shock

Citation: IMF Working Papers 2006, 200; 10.5089/9781451864601.001.A001

• Download Figure
• Download figure as PowerPoint slide

Impulse Response Functions Following a Cost-Push Shock

Citation: IMF Working Papers 2006, 200; 10.5089/9781451864601.001.A001

• Download Figure
• Download figure as PowerPoint slide

The counterfactual scenario for the pre-Volcker period in Figure 2 evaluates the significance of a change in both asset market participation and monetary policy rule for the transmission of shocks. In particular, the impulse response functions of the counterfactual scenario describe the dynamics that would have occurred in the pre-Volcker period in response to a cost-push shock, if the Fed had followed the active policy rule adopted later in the Volcker-Greenspan period, and asset market participation had been of equal degree as estimated for the Volcker-Greenspan era. Interestingly, the responses are very similar to those of the Volcker-Greenspan era, indicating that the joint changes in these two parameters explain the bulk of the change in the transmission of cost-push shocks. A similar picture emerges from the impulse response functions following an identified³⁷ “demand” shock presented in Figure 3.

Impulse Response Functions Following an Aggregate Demand Shock

Citation: IMF Working Papers 2006, 200; 10.5089/9781451864601.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Impulse Response Functions Following an Aggregate Demand Shock

Citation: IMF Working Papers 2006, 200; 10.5089/9781451864601.001.A001

• Download Figure
• Download figure as PowerPoint slide

Impulse Response Functions Following an Aggregate Demand Shock

Citation: IMF Working Papers 2006, 200; 10.5089/9781451864601.001.A001

• Download Figure
• Download figure as PowerPoint slide

V. Conclusions

The U.S. economy in the 1965−80 period was characterized by a high degree of financial regulation and limited asset-market participation; this changed in the early 1980s, owing to both deregulation and financial innovation. We reviewed institutional evidence supporting this statement and outlined a dynamic general equilibrium model incorporating limited participation in asset markets. The model predicts a change in the sign of slope of the IS curve following an exogenous structural change in asset-market participation from low to high. We showed that under such conditions (labeled “non-Keynesian”), a passive policy rule is required by (i) equilibrium determinacy, and (ii) optimal policy in the sense of maximizing welfare. Furthermore, we argued that a central bank behaving in a welfare-maximizing manner would have switched from a passive to an active policy rule endogenously in response to a change in asset-market participation. We provided empirical evidence, using Bayesian estimation techniques, that the data are compatible with our hypothesis that the sensitivity of aggregate demand to real interest rates changed sign from positive during the pre-Volcker period to negative thereafter. This sign change is triggered by the increase in asset market-participation in the Volcker-Greenspan period. Based on the foregoing theoretical and empirical results, we suggested that pre-Volcker Fed policy was better than is usually thought. Indeed, at a deep level, our results indicate that there was no change in monetary policy conduct; that is, policy was consistent throughout the whole sample with equilibrium determinacy and minimizing macroeconomic variability.

Since our framework implies that the equilibrium in the pre-Volcker period was determinate, we were able to study the effects of fundamental shocks. Stochastic simulations indicate that our model is able to replicate stylized facts of the U.S. economy for the period under scrutiny. Counterfactual simulations suggest that most, but not all, of the changes in outcomes can be explained by changes in the structure of the economy. However, the changes in shocks’ distributions are important in order to explain some key changes, such as the fall in the volatility of output. Finally, we found that theoretical responses to fundamental shocks also conform with the stylized facts. Notably, we found that cost-push shocks generated considerably higher inflation, more persistent inflation, and deeper recessions in the pre-Volcker period than they have in the Volcker-Greenspan period. Most of the change in the transmission of shocks across the two samples can be accounted for by the changes in asset-market participation and in the monetary policy rule. All in all, our results contribute to an explanation of the change in business cycles based on a change in the structure of the economy combined with a change in the distributions of fundamental shocks, rather than on “better policy” that ruled out the previously prevailing effects of sunspot shocks.

The explanation proposed here abstracts from a few aspects emphasized by others: inflation bias; and information imperfections, beliefs, and learning. This is not to say that we believe such aspects have nothing to contribute to explaining the “Great Inflation”. We merely argue that our explanation captures some features that other theories, by themselves, do not. In that sense, it could be part of the explanation, together with other, complementary and consistent theories. What weighting it should receive in solving the puzzle is, of course, an open issue.