The Time-series Linkages between US Fiscal Policy and Asset Prices

Abstract

This article studies the interplay of fiscal policy and asset price returns of the United States in a time-varying parameter vector autoregressive (VAR) model. Using annual data from 1890 to 2013, we study the effects of dynamic shocks to both fiscal policy on asset returns and asset returns on fiscal policy. Distinguishing between low-volatility (bull market) and high-volatility (bear market) regimes together with a time-varying parameter VAR model enables us to isolate the different sizes and signs of responses to shocks during different time periods. The results indicate that increases in the primary surplus-to-gross-domestic-product ratio decrease house returns over the entire sample and at each impulse horizon. Unlike the house return responses, stock returns only decrease in the first year after the fiscal shock but then increase for the following eight years. Furthermore, the findings show that asset return movements affect fiscal policy, whereby fiscal policy responds more to equity returns than to house returns. The response of fiscal policy to asset returns proves relatively stable and constant over time while controlling for various asset return regimes. Asset returns respond uniformly to fiscal policy shocks since the 1900s.

Keywords

TVP-VAR countercyclical fiscal policy stock returns house returns

Fiscal policy shocks exert wide-ranging effects, which include movements in asset returns. The policy response to the 2008/2009 Great Recession generated enormous government bailouts of “too big to fail” businesses. These bailouts produced an expansion of the US government spending, requiring either larger taxes or higher sovereign debt. The 2008/2009 Great Recession captures one outlier with respect to the interaction between asset returns and fiscal policy. These outliers, together with a changing fiscal stance over the business cycle, highlight the nonlinear relationship between asset returns and fiscal policy. We contribute to the existing literature by analyzing the simultaneous effects of fiscal policy shocks on asset returns as well as the effects of asset return shocks on fiscal policy. We depart from the existing literature that uses vector autoregressive (VAR) models that average out period effects. Agnello, Dufrenot, and Sousa (2013) provide the exception, using a time-varying transition probability (TVTP) Markov-switching (MS) framework to adjust US fiscal policy for asset prices.

This study—the first, as far as we know—analyzes the interaction of fiscal policy and asset returns for the United States in a parsimonious restricted time-varying multivariate model that controls for nonlinearities.¹ The time-varying results can capture key nonlinear effects such as explosive asset prices and various regime switches. We use the generalized sup-augmented Dickey–Fuller (GSADF) test (Phillips, Shi, and Yu 2015) to identify periods of explosive behavior in asset returns. We also implement a MS framework to identify multiple asset return regimes (bull vs. bear markets). Both tests justify the use of our nonlinear VAR framework. We use a long history of annual data from 1890 to 2013 to capture important events and regimes in the movement in asset returns and to control for changes in the fiscal stance. The results of the TVP-VAR show that both shocks to house and equity returns exert a larger effect on fiscal policy over time. Asset return shocks during the early 1900s did not generate much of an effect on fiscal policy. This outcome changed significantly since the 1970s. The response of asset returns to fiscal policy shocks depend on the duration of the shock. House returns react negatively to an increase in the fiscal primary surplus to gross domestic product (GDP) in all periods after the fiscal shock. Equity returns decrease in the initial periods after the fiscal shock. The equity return response to a fiscal policy shock, however, reverses after one year, then increasing for the following eight years.

The increase in revenue due to an increase in asset prices may create a superficially higher revenue stream (possibly procyclical). This cyclically higher revenue may create the illusion of permanently higher revenue and, as a result, may lead to higher fiscal spending. Budget balance calculations should, thus, strip the effects of asset prices on revenue to give an indication of structural revenue. Agnello, Dufrenot, and Sousa (2013) do this exactly in a TVTP-MS framework. They show that taxes adjust nonlinearly to asset prices, but spending remains neutral during an asset price cycle. This finding provides the most important justification for our contribution in a simultaneous systems model. We believe that the interaction between government taxation and spending proves important and, thus, requires study with movements in asset prices.

Literature

Afonso and Sousa (2011) show that unexpected changes in fiscal policy increase the variability in asset prices.² Aye et al. (2012) also study the effect of unexpected changes in fiscal policy on asset prices in South Africa. They show that fiscal policy announcements reduce the effects of fiscal policy shocks on asset prices in a sign-restricted VAR.

Tavares and Valkanov (2012) examine the effect of government spending and revenue on equity returns and government and corporate bond returns in the United States. An increase in government revenue to GDP significantly lowers both equity and bond returns. An increase in government spending to GDP significantly increases only bond returns. The authors discover that fiscal policy produces return variability on the same order as Federal Reserve policy. Thus, the authors conclude that fiscal policy shocks should receive more attention with respect to asset pricing.

Agnello and Sousa (2013) analyze the effect of fiscal policy in a panel VAR of ten industrialized countries. They show that a fiscal expansion exerts a negative effect on stock and house prices in the United States. Agnello, Castro, and Sousa (2012) provide a brief summary of the importance of the interaction between fiscal policy and asset prices. Two main channels exist through which asset prices affect fiscal policy: a direct channel where tax revenue increases in line with an increase in asset prices and an indirect channel where an increase in asset prices boost consumer confidence and, hence, increases aggregate demand. Eschenbach and Schuknecht (2002) show that asset prices affect fiscal balances mainly through the revenue channel. That is, capital gains, turnover-related taxes, and wealth effects affect consumption and, thus, the fiscal balance. That study, for seventeen Organization for Economic Cooperation and Development countries, shows that, on average, a 10-percent change in real estate and stock prices exert a similar effect on the fiscal balance as does a 1-percent change in output.

Liu, Mattina, and Poghosyan (2015) argue that the calculation of structural fiscal balances not only requires adjustment for the business cycle but also adjustment for asset price cycles, which do not fully synchronize with the business cycle. Thus, failure to account for asset price cycles in the calculation of the structural fiscal balance leads to a misleading signal about the underlying fiscal stance of the government. Further, asset price increases cause fiscal revenue to increase temporarily, which causes spending to respond to the higher revenue and produces a procyclical policy stance.³ Tagkalakis (2011b) reenforces this revenue channel, finding that higher asset prices positively affect the primary balance through both higher revenue and lower spending.⁴

Agnello, Castro, and Sousa (2012) use a variety of methods (linear and nonlinear) to study the response of fiscal policy to asset prices in the United States and the United Kingdom. They show that asset prices do not affect primary spending. Taxes, however, do experience a significant effect—taxes decrease when stock prices rise and increase when house prices increase. Their nonlinear model shows that during economic downswings, fiscal policy expands and offsets some of the reduction in wealth which typically occurs during recessions, while a fall in wealth associates with a tax cut.

Finally, our approach is similar to the method in Gupta, Jooste, and Matlou (2014). They estimate a TVP-VAR for South Africa and show that a nonlinear relationship exists between fiscal policy and asset prices. They also identify two significant regimes with different relationships. They demonstrate that fiscal expansions reduced both asset and house prices from the 1970s until the mid-1990s. Since 2000, however, asset prices increased in response to fiscal expansions. One explanation argues that fiscal policy during the 2000 to 2010 period was conducted in a sustainable and countercyclical way, prompting consumers and firms to trust that the fiscal policy stance will stimulate aggregate demand during recessions and to not during economic expansions. Increasing taxes could limit the consolidation process, especially when it reduces asset prices (see Aye et al. 2012). Considering the reverse effect from asset prices to fiscal policy, as expected, an increase in asset prices reduced deficits.

Empirical Methodology

To motivate the TVP-VAR model, we test for nonlinearities in asset price movements. We use two methods for this purpose: a GSADF test and an MS regression. The GSADF tests for bubbles and identifies the time origin of bubbles (Phillips, Shi, and Yu 2015). The test can detect bubbles in long time-series data. The GSADF test is a recursive right-tailed augmented Dickey–Fuller (ADF) test with flexible windows, where both the start and end points change.⁵ We also use Markov-switching autoregression (MS-AR) to complement the GSADF test. The MS-AR regressions identify the various bull and bear regimes.

In sum, we use the MS-AR models and the bubble tests to identify bear and bull markets and possible periods of irrational exuberance in the two asset markets independent of any other variables. Then, we relate the time-varying impulse responses of the TVP-VAR model over these periods to identify any specific patterns in the impulse responses over the identified periods of bear or bull markets and bubbles. That is, do the impulse responses for a shock to budget deficit as a percentage of GDP (asset prices) on the asset prices (budget deficit as a percentage of GDP) differ under bear or bull markets and during bubbles?

Once we verify the presence of nonlinearities and identify periods of asset return exuberance, we use a TVP-VAR to analyze shocks over time and over the identified regimes. Time-varying impulses allow us to study the evolution of how fiscal policy shocks affect asset returns and vice versa. Time variation comes from both the parameters and the variance covariance matrix of the model’s innovations. This reflects simultaneous relations among variables of the model and heteroscedasticity of the innovations (Primiceri 2005). To accomplish this, we use a Markov chain Monte Carlo (MCMC) algorithm to estimate the coefficients and the multivariate stochastic volatility. Researchers commonly estimate time variation (see Sims 1993; Stock and Watson 1996; Cogley and Sargent 2001). These studies, however, impose restrictions on the variance covariance matrix that should evolve over time. Most of these models limit themselves to reduced-form models that can only describe and forecast data (Primiceri 2005). With drifting coefficients, one essentially captures the learning process and possible nonlinearities or time variation in the lag structure of the model. The multivariate stochastic volatility can capture possible heteroscedasticity of the shocks and nonlinearities in the simultaneous relations among the variables of the model.

Following Nakajima (2011), this article estimates a time-varying parameter VAR model with stochastic volatility of the form:

y_{t} = c_{t} + B_{1 t} y_{t - 1} + ... + B_{s t} y_{t - s} + e_{t}, e_{t} \sim N (0, Ω_{t}),

where $t = s + 1, . . ., n,$ y_t is a $(k \times 1)$ vector of observed variables, $B_{1 t}, \dots, B_{s t}$ are $(k \times k)$ matrices of time-varying coefficients, and $Ω_{t}$ is a $(k \times k)$ time-varying covariance matrix. We assume a recursive identification scheme through the decomposition of $Ω_{t} = A_{t}^{-1} Σ_{t} Σ_{t} A_{t}^{' -1}$ , where A_t is a lower-triangle matrix with diagonal elements equal to one and $Σ_{t} = diag (σ_{1 t}, . . ., σ_{κ t})$ . We define $β_{t}$ as the stacked row vector of $B_{1 t}, \dots, B_{s t}$ ; α _t is the stacked row vector of the free lower-triangular elements of A_t ; and $h_{t} = (h_{1 t}, ...., h_{k t})$ , where $h_{j t} = log σ_{j t}^{2}$ . We assume that the time-varying parameters follow a random-walk process. Thus,

\begin{array}{l} β_{t + 1} = β_{t} + υ_{β t}, \\ α_{t + 1} = α_{t} + υ_{α t}, \\ h_{t + 1} = h_{t} + υ_{h t}, \end{array} (\begin{array}{l} ε_{t} \\ υ_{β t} \\ υ_{α t} \\ υ_{h t} \end{array}) \sim N (0, (\begin{matrix} I & 0 & 0 & 0 \\ 0 & Σ_{β} & 0 & 0 \\ 0 & 0 & Σ_{α} & 0 \\ 0 & 0 & 0 & Σ_{h} \end{matrix})),

where $t = s + 1, . . . , n,$ $e_{t} = A_{t}^{- 1} Σ_{t} ε_{t}$ , $Σ_{α}$ and $Σ_{h}$ are diagonal, $β_{s + 1} ∼ N (μ_{β o}, Σ_{β o}),$ $α_{s + 1} ∼ N (μ_{α o}, Σ_{α o}),$ and $h_{s + 1} ∼ N (μ_{h o}, Σ_{h o})$ .⁶ We use Bayesian inference to estimate the TVP-VAR model via MCMC methods, which assesses the joint posterior distributions of the parameters of interest under certain prior probability densities. We assume the following priors, as in Nakajima (2011): $Σ_{β} \sim I W (25, 0.01 I)$ , ${(Σ_{α})}_{i}^{- 2} \sim G (4, 0 .02),$ and ${(Σ_{h})}_{i}^{- 2} \sim G (4, 0.02),$ where ${(Σ_{α})}_{i}^{- 2}$ and ${(Σ_{h})}_{i}^{- 2}$ are the ith diagonal elements in $Σ_{α}$ and $Σ_{h}$ , respectively, and $I W$ and G denote the inverse Wishart and gamma distributions, respectively. For the initial set of time-varying parameters, we set flat priors such that $μ_{β o}$ = $μ_{α o}$ = $μ_{h o}$ =0 and $Σ_{β o}$ = $Σ_{α o}$ = $Σ_{h o} = 10 \times I .$

The VAR exhibits a recursive identification scheme, where depending on the ordering of the variables, contemporaneous shocks are zero. We order the variables from most exogenous to least exogenous with house returns first, followed by the primary surplus, and then equity returns.⁷ This ordering is theoretically justified (Gupta, Jooste, and Matlou 2014), since it implies that the house return does not respond contemporaneously to fiscal policy and equity return shocks, while fiscal policy reacts with a lag to equity return shocks. Thus, the equity return appears third in the ordering after the house return and our measure of fiscal policy.

Data Description

Our variables include the real US house price (RHP) index, real Standard and Poor’s S&P500 (RSP) index, and the US primary surplus relative to GDP (BB).⁸ We use annual data from 1890 to 2013 because data only occur at this frequency over this long sample. We use yearly growth rates (log-differences) for the real house and stock returns (RHR and RSR, respectively), which generates 123 observations, covering 1891 to 2013.

We plot the RHP and RSP series in figure A1 in the Appendix. Both series exhibit nonstationary behavior at the conventional 5-percent significance level. The level of BB and the transformation of the two asset prices into returns generate stationary series based on standard unit root tests (i.e., ADF; Dickey and Fuller 1981), Phillips–Perron (PP; Phillips and Perron 1988), Dickey–Fuller with generalized least squares detrending (Elliott, Rothenberg, and Stock 1996), and the Ng–Perron modified version of the PP (Ng and Perron 2001), which table A1 in the Appendix reports the results. We standardize all variables to ensure that we can easily compare the results across the two asset returns. Summary statistics of the nonstandardized variables, however, appear in table A2 in the Appendix. All variables exhibit nonnormal behavior, providing initial motivation to include stochastic volatility in the model.

Our univariate MS regressions identify two regimes for both house and equity returns: a high-volatility state (regime 1), which is a bear market, and a low-volatility state (regime 2), which is a bull market (see Simo-Kenge, Miller, and Gupta [2016] and references therein). Table 1 summarizes the results of the MS-AR regressions. The observed volatility is higher for both house and equity returns during bear markets as compared to bull markets. The expected duration in the bear market for house returns is about 80 years compared to about 34 years during the bull market.⁹ For equity returns, the duration for bear and bull markets proves much shorter at about 2 and 2.5 years, respectively.

Table 1.

Markov-switching Autoregressive (MS-AR) Estimates.

Parameters	Real House Returns(MS-AR[3])	Real Stock Returns(MS-AR[2])
Regime-dependent intercept
$μ_{1}$	−0.174	−7.204
$μ_{2}$	0.388	10.710***
Standard errors
$σ_{1}$	2.143***	2.855***
$σ_{2}$	0.710***	2.423***
Regime 1: Parameter estimates
$β_{1, 1}$	0.008	0.275
$β_{1, 2}$	0.018	−0.431**
$β_{1, 3}$	−0.168
Regime 2: Parameter estimates
$β_{2, 1}$	1.027***	−0.189
$β_{2, 2}$	−0.171	0.035
$β_{2, 3}$	−0.165
Fit
Log-likelihood	−364.256	−514.627
Transition probabilities
Bear	0.988	0.549
Bull	0.971	0.606
Expected duration
Bear	80.328	2.219
Bull	34.806	2.541

* Indicates significance at 10 percent.

** Indicates significance at 5 percent.

*** Indicates significance at 1 percent.

Figure 1 plots the regime probabilities for a bull market for both stock and house returns. Stock and house return bull markets nearly coincide with each other prior to 1950, which does not occur in the post-1950 period. The housing market experienced a long low-volatility period from 1950 until the collapse of housing market leading up to the financial crisis in 2008.

Figure 1.

Markov-switching autoregressive bull market probabilities for real stock and house returns.

Similar to Pavlidis et al. (2013), the GSADF test for the real house price shows signs of explosive behavior during the early 2000s until around 2007 (figure 2). The correction from this bubble helps to explain the 2008/2009 Great Recession. A real stock price bubble (figure 3) emerges from 1995 to 2003. Phillips, Shi, and Yu (2015) use a higher-frequency series for the S&P data over a similar time period and identify a bubble from 1995 to 2008. Both asset prices in real terms reveal an explosive and unsustainable bubble that preceded the Great Recessions.

Figure 2.

Generalized sup-augmented Dickey–Fuller test for house returns.

Figure 3.

Generalized sup-augmented Dickey–Fuller test for equity returns.

Results

Having identified key periods of asset return exuberance and potential bubbles, we turn to the TVP-VAR model to analyze the response of asset returns to fiscal policy shocks as well as the response of fiscal policy to asset return shocks.¹⁰

We generate the posterior estimates after drawing 10,000 samples, with the first 1,000 draws discarded. Table 2 reports these posterior estimates for the means, along with those for the standard deviations, the 95-percent credibility intervals,¹¹ the convergence diagnostic (CD) due to Geweke (1992), and the inefficiency factors.¹² The 95-percent credibility intervals include the estimates for the posterior means, and the CD statistics do not allow us to reject a null hypothesis of convergence to the posterior distribution at the 5-percent significance level. Furthermore, the inefficiency factors are relatively low. We can, thus, conclude that the MCMC algorithm efficiently produces the posterior draws. This, in turn, also implies that even with annual data (instead of the traditional quarterly data used in this literature), our model does not suffer from imprecision in the parameter estimates and the generated impulse responses (which we discuss below).

Table 2.

Selected Estimation Results.

Parameter	Mean	St. Dev.	95 Percent Confidence Intervals	CD	Inefficiency
${(\sum_{β})}_{1}$	.0023	.0003	[0.0018, 0.0029]	.018	5.68
${(\sum_{β})}_{2}$	.0023	.0003	[0.0018, 0.0029]	.074	7.76
${(\sum_{α})}_{1}$	.0056	.0016	[0.0034, 0.0095]	.899	34.23
${(\sum_{α})}_{2}$	.0054	.0014	[0.0033, 0.0088]	.628	23.73
${(\sum_{h})}_{1}$	.3539	.0960	[0.1950, 0.5580]	.089	35.15
${(\sum_{h})}_{2}$	.8358	.1542	[0.5611, 1.1567]	.164	14.82

Note: CD = convergence diagnostic.

Figure 4 plots the posterior estimates of stochastic volatility for each of the variables used in the TVP-VAR. The top row plots the actual time series, while the bottom row plots the posterior estimates of the stochastic volatility of each series. The stochastic volatility of equity returns remains more or less constant, but high, throughout our sample. The most prominent point in the posterior estimate of the primary surplus stochastic volatility occurs during World War II (1939–1950), which matches what we observe in the evolution of fiscal policy during that period. Smaller movements in volatility also occur for World War I, and even smaller, but noticeable, movements occur during the Great Recession. Stochastic volatility for house returns achieved its highest level at the end of the nineteenth century and increased again from 2000 to 2007. The stochastic volatility between fiscal policy and asset returns does correspond over specific periods. In contrast, Simo-Kenge, Miller, and Gupta (2016) show that monetary policy and asset return volatility correlate over specific periods. The higher volatility in house returns during the beginning of the 1900s to around 1940 corresponded to relative low volatility in interest rates. The nonconstant posterior estimates for stochastic volatility justify our use of the model with stochastic volatility as opposed to one with constant volatility in the structural shocks.

Figure 4.

Posterior estimates for the stochastic volatility of the structural shock.

Figure 5 shows that both house and equity returns respond negatively to a fiscal surplus shock contemporaneously, a finding similar to one in Gupta, Jooste, and Matlou (2014). We have nine impulse horizons from 1892 to 2013. We summarize, however, only impulses at years 1, 3, 6, and 9. The response of house returns to a fiscal shock lasts up to three years, whereas the response of stock returns to a fiscal policy shock lasts only one year. This is similar to figure A3 (see Appendix) for a constant parameter VAR, although the real stock return increases initially for a period before turning negative as well, opposite to what the TVP-VAR model generates.

Figure 5.

Median response of asset returns to a primary surplus shock.

This negative response to fiscal shocks reflects a standard market response. That is, when the primary surplus increases, or more likely the primary deficit decreases, the supply of government debt falls, driving up the price of government debt and lowering its interest rate. As such, the quantity demanded of government debt declines, moving to the housing and equity markets, lowering the returns in these markets. Interestingly, we see little volatility in the impulse responses over time.

The impulse responses of figure 6 ¹³ show how the primary balance reacts to asset return shocks over time. Both house and equity return shocks lead to higher primary surpluses over time. This matches the constant parameter VAR of figure A4 for the first couple of years (see Appendix). We expect increases in house and equity returns, a priori, to increase automatically the revenue collected from these tax bases. At the same time, we also expect countercyclical spending from government during periods of buoyant growth. Interestingly, the effects of asset return shocks on the primary balance exhibit stability since the 1990s. This indicates that the government did not react differently to asset return shocks during two significant periods of asset price movements: the dot com bubble in 2000 and the 2008/2009 Great Recession.

The primary balance reacts more to stock return shocks contemporaneously compared to house return shocks, indicating that movements in equity markets provide either a stronger source of revenue or a signal to government of changes to the economic cycle. Of course, tax revenue changes occur only for assets sales, realized gains, or losses. Since houses contain both consumption and investment components, whereas equities include only the investment component, realized gains or losses will occur more frequently for equities than for houses. Nonetheless, the effect of stock return shocks on the primary balance quickly dies out after one period.

These responses increase smoothly, although at differing rates, over time (see figures 5 and 6). The exception is that the real stock return response to fiscal shocks remains relatively uniform since the 1900s, albeit at different levels (see figure 5).

Figure 6.

Median response of the primary balance to asset return shocks.

Finally, we performed the MS and bubble tests on the univariate data to see if the behavior of asset prices and fiscal policy differed somewhat between bear and bull markets as well as during bubbles in the two asset markets. The smoothness of the impulses in figures 5 and 6 suggest that the impulses do not differ dramatically between bear and bull markets and/or during bubbles.

Robustness Tests

Based on two anonymous referees’ recommendation, we conducted the following robustness tests. First, instead of the three-variable VAR, we used a four-variable VAR, where we included the real GDP growth rate along with the real budget surplus. Following Agnello and Sousa (2013), we order the variables as RHR, real GDP growth, BB, and RSR. The results from the inclusion of the real GDP growth rate appear in figures A9 and A10 and qualitatively match those of figures 5 and 6, including the relative magnitude of the findings for the 1-, 3-, 6-, and 9-year impulse horizons.

Second, also in the benchmark model with three variables, we added the real interest rate (the nominal interest rate minus the CPI inflation rate, obtained from the data segment of professor Robert J. Shiller’s website). Following Agnello and Sousa (2013), we ordered the variables as RHR, BB, real interest rate, and RSR. Once again, the findings shown in figures A11 and A12 qualitatively match those of figures 5 and 6, including the relative magnitudes of the findings for the 1-, 3-, 6-, and 9-year impulse horizons.

Finally, we replaced the budget-surplus-to-GDP measure of fiscal policy with tax-revenue- and government-expenditure-to-GDP ratios and conducted the analysis over the period of 1929 to 2013 (due to the data availability from the US Census Bureau). That is, the data on government spending and tax revenue do not exist separately for the entire period that we consider. Hence, we could not disaggregate the primary deficit into these components for the full sample. Thus, we conducted this analysis only for 1929 to 2013, where data do exist on the separate components. Our results, once again, qualitatively match those in figures 5 and 6 for the tax-revenue-to-GDP analysis shown in figures A13 and A14. The findings invert when we use the government-expenditure-to-GDP ratio because government expenditure enters the budget surplus variable with a negative sign. Nevertheless, the findings for the government expenditure variable after adjusting for the sign shown in figures A15 and A16 qualitatively match those of figures 5 and 6, including the relative magnitudes of the findings for the 1-, 3-, 6-, and 9-year impulse horizons.

Further, we found that the tax-revenue-to-GDP ratio decreased asset returns more strongly than the increase in asset returns resulting from an increases in the government-expenditure-to-GDP ratio. In addition, an increase in asset returns also increased the tax-revenue-to-GDP ratio relatively more than the reduction in the government-expenditure-to-GDP ratio. This explains why we see a fall in asset returns following an increase in the budget-surplus-to-GDP ratio and an increase in the budget-surplus-to-GDP ratio due to increases in asset returns when the analysis was conducted in the three-variable VAR over 1929 to 2013.

Conclusion

This article studies the dynamic interaction between fiscal policy and asset returns in a TVP-VAR setup. This method addresses potential nonlinearities between fiscal policy and asset returns and controls for exuberant periods of asset returns as well as a changing fiscal policy stance. The use of the TVP-VAR is motivated by two tests—a GSADF tests that detects and dates bubbles and an MS regression that identifies multiple asset return regimes.

After controlling for time-varying stochastic volatility, the results show that positive asset return shocks increase the primary surplus. The response of fiscal policy to asset returns increases every year from the 1890s into the early 1990s. Fiscal policy’s reaction to asset return movements becomes fairly stable after 1990. The findings also show that fiscal policy did not overreact to drastic changes in asset returns during the 2008/2009 financial crisis.

The results also show that asset returns react negatively to an increase in the primary surplus. Stock returns, however, increase in response to a fiscal shock after a year. The TVP impulse responses remain fairly constant over time. This suggests a constant reaction of asset returns to fiscal policy shocks since 1900. Although house returns decrease due to an expansion in fiscal policy, stock returns only decrease during the first year after the shock.

We conclude by discussing anticipation effects and, hence, nonfundamentalness, which arise when the information set of the econometrician is smaller than that of economic agents (Lippi and Reichlin 1994). Given this, Ramey (2011) emphasizes that neglecting anticipation effects in fiscal VARs can cause biased impulse responses and, hence, suggests the need to include news about future fiscal policy to overcome this problem as done in Berg (2015), for instance, using professional forecasts. We cannot, however, accommodate nonfundamentalness in our study, as professional forecasts on the fiscal policy variables and asset returns do not exist over our historical sample of 1891 to 2013. This illustrates the trade-off involved in examining the historical evolution, instead of recent periods only, of the relationship between fiscal policy and asset returns.

Footnotes

Appendix

Table A2.

Summary Statistics.

Statistic	BB	RHR	RSR
Mean	−2.2899	0.2108	1.7949
Median	−0.7508	0.0728	2.6222
Maximum	4.2879	29.5108	38.1463
Minimum	−27.4693	−21.4550	−54.9023
Std. dev.	4.5299	7.0940	18.3697
Skewness	−3.0551	0.3749	−0.6145
Kurtosis	14.5459	5.3887	3.3993
Jarque–Bera	874.5433	32.1244	8.5569
Probability	.0000	.0000	.0139
Observations	123	123	123

Note: Probability relates to the Jarque–Bera test which has a null of normality. Std. dev. = standard deviation; BB = primary surplus relative to gross domestic product; RHR = real house returns; RSR = real stock returns.

Acknowledgment

We thank two anonymous referees for many helpful comments.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

References

Afonso

Sousa

R. M.

. 2011. “What Are the Effects of Fiscal Policy on Asset Markets?” Economic Modelling 28:1871–90.

Agnello

Castro

Sousa

R. M.

. 2012. “How Does Fiscal Policy React to Wealth Decomposition and Asset Prices?” Journal of Macroeconomics 34:874–90.

Agnello

Dufrenot

Sousa

R. M.

. 2013. “Using Time-varying Transition Probabilities in Markov Switching Processes to Adjust US Fiscal Policy for Asset Prices.” Economic Modelling 34:25–36.

Agnello

Sousa

R. M.

. 2013. “Fiscal Policy and Asset Prices.” Bulletin of Economic Research 65:154–77.

Auerbach

A. J.

Gorodnichenko

. 2012. “Measuring the Output Responses to Fiscal Policy.” American Economic Journal: Economic Policy 4:1–27.

Aye

Balcilar

Gupta

Jooste

Miller

S. M.

Ozdemir

. 2012. “Fiscal Policy Shocks and the Dynamic Asset Prices. The South African Experience.” Public Finance Review 42:511–31.

Bachmann

Sims

E. R.

. 2012. “Confidence and the Transmission of Government Spending Shocks.” Journal of Monetary Economics 59:235–49.

Bai

Perron

. 2003. “Computation and Analysis of Multiple Structural Change Models.” Journal of Applied Econometrics 18:1–22.

Beetsma

Giuliodori

. 2011. “The Effects of Government Purchases Shocks: Review and Estimates for the EU.” Economic Journal 121:F4–32.

10.

Berg

T. O.

2015. “Time Varying Fiscal Multipliers in Germany.” Review of Economics 66:13–46.

11.

Blanchard

Perotti

. 2002. “An Empirical Characterization of the Dynamic Effects of Changes in Government Spending and Taxes on Output.” The Quarterly Journal of Economics 117:1329–68.

12.

Born

Muller

Pfeifer

. 2016. “Does Austerity Pay Off?,” Barcelona Graduate School of Economics, Universitat Pompeu Fabra (UPF) and Universitat Autonoma de Barcelona (UAB), Spain.

13.

Cogley

Sargent

. 2001. “Evolving Post-world War II U.S. Inflation Dynamics.” NBER Macroeconomics Annual, pp. 331–73. Cambridge, MA: MIT Press.

14.

Croce

M. M.

Kung

Nguyen

Schmid

. 2012. “Fiscal Policy and Asset Prices.” Review of Financial Studies 25:2635–72.

15.

Croce

M. M.

Nguyen

Schmid

. 2012. “The Market Price of Fiscal Uncertainty.” Journal of Monetary Economics 59:401–16.

16.

Curatola

Donadelli

Gioffré

Gruning

. 2015. “Austerity, Fiscal Uncertainty, and Economic Growth: Insights from Fiscally Weak EU Countries.” SAFE Working Paper No. 56. Germany: Leibniz Institute for Financial Research.

17.

Dickey

Fuller

. 1981. “Likelihood Ratio Statistics for Autoregressive Time Series with a Unit Root.” Econometrica 49:1057–72.

18.

Elliott

Rothenberg

T. J.

Stock

J. H.

. 1996. “Efficient Tests for an Autoregressive Unit Root.” Econometrica 64:813–36.

19.

Eschenbach

Schuknecht

. 2002. “Asset Prices and Fiscal Balances.” European Central Bank, Working Paper No. 141, Frankfurt, Germany.

20.

Geweke

1992. “Evaluating the Accuracy of Sampling-based Approaches to the Calculation of Posterior Moments. In Bayesian Statistics, edited by Bernado

J. M.

Berger

J. O.

Dawid

A. P.

Smith

A. F. M.

, vol. 4, 169–88. New York: Oxford University Press.

21.

Gupta

Jooste

Matlou

. 2014. “A Time Varying Approach to Analyzing Fiscal Policy and Asset Prices in South Africa.” Journal of Financial Economic Policy 6:46–63.

22.

Ilzetzki

Mendoza

Végh

. 2013. “How Big (Small?) Are Fiscal Multipliers?” Journal of Monetary Economics 60:239–54.

23.

Jaeger

Schuknecht

. 2007. “Boom-bust Phases in Asset Prices and Fiscal Policy Behavior.” Emerging Markets Finance & Trade 43:45–66.

24.

Kirchner

Cimadomo

Hauptmeier

. 2010. “Transmission of Government Spending Shocks in the Euro Area: Time Variation and Driving Force.” European Central Bank Working Paper No. 1219, Frankfurt, Germany.

25.

Lippi

Reichlin

. 1994. “VAR Analysis, Nonfundamental Representations, Blaschke Matrices.” Journal of Econometrics 63:307–25.

26.

Liu

E. X.

Mattina

Poghosyan

. 2015. “Correcting ‘Beyond the Cycle’: Accounting for Asset Prices in Structural Fiscal Balances.” International Monetary Fund Working Paper No. 109, Washington, D.C.

27.

Morris

Schuknecht

. 2007. “Structural Balances and Revenue Windfalls: The Role of Asset Prices.” European Central Bank Working Paper No. 737, Frankfurt, Germany.

28.

Nakajima

2011. “Time-varying Parameter VAR Model with Stochastic Volatility: An Overview of Methodology and Empirical Applications.” Monetary and Economic Studies 29:107–42.

29.

Perron

. 2001. “Lag Length Selection and the Construction of Unit Root Tests with Good Size and Power.” Econometrica 69:1519–54.

30.

Pavlidis

Yusopove

Paya

Peel

Martinez-Garcia

Mack

Grossman

. 2013. “Monitoring Housing Markets for Episodes of Exuberance: An Application of the Phillips et al. (2012, 2013) GSADF test on the Dallas Fed International House Price Database.” Federal Reserve Bank of Dallas Working paper No. 165, Dallas, Texas USA.

31.

Phillips

P. C. B.

Perron

. 1988. “Testing for a Unit Root in Time Series Regression.” Biometrika 75:335–46.

32.

Phillips

P. C. B.

Shi

. 2015. “Testing for Multiple Bubbles: Historical Episodes of Exuberance and Collapse in the S&P500.” International Economic Review 56:1043–78.

33.

Primiceri

2005. “Time Varying Structural Vector Autoregressions and Monetary Policy.” The Review of Economic Studies 72:821–52.

34.

Ramey

V. A.

2011. “Identifying Government Spending Shocks: It’s All in the Timing.” Quarterly Journal of Economics 126:1–50.

35.

Simo-Kenge

B. D.

Miller

S. M.

Gupta

. 2016. “Evolution of Monetary Policy in the US: The Role of Asset Prices.” Journal of Real Estate Finance and Economics 52:226–43.

36.

Sims

C. A.

1993. “A 9 Variable Probabilistic Macroeconomic Forecasting Model.” In Business Cycles, Indicators and Forecasting: NBER Studies in Business Cycles, edited by Stock

J. H.

Watson

M. W.

, vol. 28, 179–214. Chicago: University of Chicago Press.

37.

Stock

J. H.

Watson

M. W.

. 1996. “Evidence on Structural Instability in Macroeconomic Time Series Relations.” Journal of Business and Economic Statistics 14:11–30.

38.

Tagkalakis

2011a. “Asset Price Volatility and Government Revenue.” Economic Modelling 28:2532–43.

39.

Tagkalakis

2011b. “Fiscal Policy and Financial Market Movements.” Journal of Banking and Finance 35:231–51.

40.

Tavares

Valkanov

. 2012. “Fiscal Policy and Asset Returns.” Mimeo, University of California, Los Angeles.