Evaluating the Importance of Monetary Policy Uncertainty: The Long- and Short-Term Effects and Responses

Abstract

Monetary policy changes have an irreplaceable impact on economic activity. Considering the close linkage among economic policies, we employ a bi-directional Granger causality test to investigate the potential linkages between monetary policy uncertainty (MPU) and other categorical economic policy uncertainty (CEPU) in the time and frequency domains. We consider all news-based U.S. categorical economic policy uncertainty indices (CEPU). All monthly CEPU indicators, covering January 1986 to January 2022, can be obtained from the website of Economic Policy Uncertainty. On an average, causality running from each CEPU to MPU is not apparent, while MPU can significantly affect six policy-related uncertainties: taxes, government spending, health care, national security, entitlement programs and regulation. A further frequency-domain study showed the dynamic changes in the relationship between them. For instance, we capture mid- and long-run causality running from tax uncertainty to MPU, while MPU has an impact on taxes in the medium run. Our findings provide policymakers with a better understanding of the nexus between MPU and other CEPU for formulating appropriate economic policies. Particularly, if a sectional government considers the long- and short-term effects of different policies when formulating strategies, risk transmission may be curbed to some extent.

Keywords

monetary policy uncertainty categorical economic policy uncertainty Granger causality test time domain frequency domain

Introduction

The importance of monetary policy cannot be overemphasised. However, it is worth mentioning that monetary policy implementation is always accompanied by uncertainty, which could bring about some risks for policymakers and investors (Jordà & Salyer, 2003; Li et al., 2022). For example, in anticipatory planning, central banks may become passive in formulating policies because of unexpected macroeconomic changes, which will cause considerable uncertainty to the financial market (Ugurlu-Yildirim et al., 2021). In addition, the timing of policy decisions and their potential impact, in particular, have a greater impact on investors and financial institutions, when they adjust their portfolios (Kurov & Stan, 2018; Liang et al., 2022). Therefore, monetary policy uncertainty leads to a series of movements in financial markets or other aspects, which will further affect other policy implementations, especially in certain extreme circumstances (Friedman, 2015). For instance, the ongoing conflict between Russia and Ukraine has increased uncertainty of monetary policy, further affecting fiscal policy in Europe, and hampering economic activity in the EU in several ways (Liadze et al., 2022). Hence, to promote economic recovery and maintain financial stability during these extreme events, scholars have focused on monetary policy uncertainty (Azad & Serletis, 2021; Kurov & Stan, 2018; Tillmann, 2020).

Several studies have investigated the links between monetary policymaking and different economic or non-economic factors such as stock price volatility, climate change, trade policy and investor behaviour (Cacciatore & Ghironi, 2021; Mukherjee & Ouattara, 2021; Rigobon & Sack, 2003). Thorbecke (1997) examines how monetary policy shocks affect stock returns. Kurov (2010) explored the responses of investor sentiment and stocks to monetary policy decisions. Yagihashi and Du (2015) construct a general equilibrium model to reveal how healthcare reacts to expansionary monetary policy shocks. Hence, monetary policy appears irreplaceable in maintaining the stability of the financial system. It should be noted that Baker et al. (2016) pointed out close relationships among different economic policies. Additionally, unlike previous studies, they focus further on the uncertainty generated by different economic policy changes, and then construct a novel branch of indices that can capture these categorical economic policy uncertainties (CEPUs), including monetary policy uncertainty (MPU). Subsequently, this EPU series has received increased attention (Al-Thaqeb & Algharabali, 2019; Bonaime et al., 2018; Guo et al., 2022; Liang et al., 2022). Based on their contributions and the consideration of the important role of monetary policy, we cannot pay attention only to the uncertainty originating from monetary policy and other categorical economic policies, but also try our best to fully understand the relationship between MPU and CEPUs. Hence, in this study, we investigate the causal relationships between MPU and these CEPUs to enrich the existing research. In addition, we focus on exploring whether these linkages are permanent or transient. This understanding may provide reference suggestions for portfolio optimisation and risk management from a retrospective perspective. We contribute to the current research in three ways.

First, we systematically analyse the relationship between MPU and all news-based CEPUs. A large number of studies have focused on economic policy uncertainty, and verified the importance of these indices in the context of financial and economic analysis. For instance, Bonaime et al. (2018) pointed out that the acquisition likelihood could be negatively affected by CEPUs, including uncertainties in monetary policy, fiscal policy and regulation. However, they did not explore whether other policy uncertainties could impact the acquisition likelihood. Gabauer and Gupta (2018) detected economic policy uncertainty spillovers between the U.S. and Japan and found that MPU is the most dominant factor. However, they did not aim to detect the relationship between the different CEPUs. Guo et al. (2022) investigated the connectedness among only three types of CEPUs, not taking all of them into consideration. In other words, scholars have only focused on the relationship between certain economic policy-related uncertainty indicators and a specific variable, while ignoring the differences in the impact of different policy uncertainties on this variable. Therefore, this study fills the existing research gap from a holistic perspective.

Second, we employ the Granger causality test proposed by Granger (1969) to explore the linkages between them. Granger causality test, a popular tool that can detect the risk transmission between financial fundamentals, has been receiving a lot of attention (Hong et al., 2009; Peng et al., 2018; Zhang et al., 2014). In addition, this bi-directional testing method is easy to implement and has been applied in many areas such as political relationships, neuroscience, terrorism and environmental research (Freeman, 1983; Lopez & Weber, 2017; Seth et al., 2015). More importantly, it enables us to detect the impact and response of MPU on CEPUs simultaneously. Al-Thaqeb and Algharabali (2019) pointed out that overall economic policy uncertainty (EPU)¹ can affect monetary policy changes. However, they did not focus on whether monetary policy could impact EPU. A large number of works pointed out that MPU plays a significant role through its wide influence on different areas (De Pooter et al., 2021; Gupta & Jooste, 2018; Kurov & Stan, 2018). Hence, we enrich the current literature by not only detecting the impacts, but also the responses of MPU to other CEPUs.

Finally, to observe the long- and short-term characteristics of this causality, we apply the frequency-domain Granger causality test. Strohsal et al. (2019) pointed out that the conventional time-domain Granger causality method fails to capture some significant relationships that can exist only at specific frequencies. In other words, frequency-domain results can provide causal relationships over different periods, especially in the long and short term (Bodart & Candelon, 2009). Economically, some policy formulations and investments may concentrate on short-term effects, whereas others focus on long-term effects. Based on this consideration, Breitung and Candelon (2006) extended the time-domain Granger causality test from a frequency-domain perspective, which is convenient for researchers, to assess whether the linkage between variables is permanent or short-lived (Bozoklu & Yilanci, 2013; Joseph et al., 2014; Yamada & Yanfeng, 2014). For instance, Tiwari (2012) examined the long- and short-run relationship between producer and consumer prices, and Croux and Reusens (2013) investigated the predictive power of stock prices on GDP in G-7 countries. Therefore, considering that the strength or direction of the Granger causality between MPU and CEPUs will change over frequencies, we apply a frequency-domain Granger causality test. Many studies have explored the relationship between CEPUs while overlooking investigating their long- and short-term influences. For instance, Thiem (2018) analysed the cross-category spillovers between these CEPUs but ignored their short- and long-term dynamics. Although Guo et al. (2022) revealed the spillover effects among three CEPUs by employing the frequency-domain decomposition approach, they did not investigate the other CEPUs in detail. Thus, we extend our work further from a frequency-domain perspective.

The remainder of this paper is organised as follows. The Methodology section introduces the study methodology. The Data section describes the data used in this study. In the Empirical Findings section, we conduct an empirical analysis and discuss the results. We then propose policy implications in the Discussion and Suggestions section. The Other Uncertainty Indices section supplements studies on other uncertainty indices. Finally, the Conclusion section concludes the paper.

Methodology

Granger Causality Test

Granger causality was first introduced by Granger (1969) in the field of economics; then Sims (1972) proposed a method that tests the Granger causality of two time series, based on the VAR model. Specifically, if the past values of $X$ include useful information for forecasting $Y$ , $X$ can Granger-cause $Y$ ( $X \to Y$ ). In contrast, $Y$ Granger-causes $X$ , namely, $Y \to X$ . The coexistence of $X \to Y$ and $Y \to X$ indicates a bi-directional Granger causality between X and Y.

Taking the null hypothesis of $Y_{t} ↛ X_{t}$ as an illustration, it can be tested in the following ways. Assume that

X_{t} = X_{0} + \sum_{i = 1}^{m} a_{i} Y_{t - i} + \sum_{i = 1}^{m} b_{i} X_{t - i} + μ_{t}

(1)

where

{X_{t}}

and

{Y_{t}}

are stationary,

X_{0}

is a constant,

μ_{t} \sim N I D (0,1)

Y_{t - i}

and

X_{t - i}

are the

t - i

lagged values of

Y

and

X

, respectively. For each

i (1, \dots, m)

, the null hypothesis is equal to

a_{i} = 0

. Considering the limitations of F-statistics, we test the null hypothesis using Wald statistics (Lütkepohl & Reimers, 1992).²

Frequency-Domain Granger Causality Test

The strength or direction of Granger causality may change over time (Breitung & Candelon, 2006). It was assumed that $Z_{t} = {(X_{t}, Y_{t})}^{T}$ , and

Θ (L) Z_{t} = μ_{t}

(2)

where

Θ (L) = (\begin{array}{c} θ_{11} (L) & θ_{12} (L) \\ θ_{21} (L) & θ_{22} (L) \end{array})

θ_{i j} (L), i, j = 1, 2

is an autoregressive polynomial with lag operator

L

μ_{t} = (\begin{array}{c} μ_{1 t} \\ μ_{2 t} \end{array})

μ_{i t} (i = 1, 2)

is white noise with

E = (u_{t} u_{t}^{'}) = Σ

. Then, we can deduce

Z_{t} = ψ (L) η_{t}

(3)

where

ψ (L) = Θ {(L)}^{- 1} G^{- 1}

G G^{T} = Σ^{- 1}

and

η_{t} = G ε_{t}

with

E (η_{t} η_{t}^{'}) = I

. Then, the spectral density of

X_{t}

f_{X_{t}} (ω) = \frac{1}{2 π} {{| Ψ_{11} (e^{- i ω}) |}^{2} + {| Ψ_{12} (e^{- i ω}) |}^{2}}

(4)

where

ω

denotes frequency. According to Geweke (1982), the causality from

X_{t}

Y_{t}

is defined as

M_{Y_{t} \to X_{t}} (ω) = \log [\frac{2 π f_{x} (ω)}{{| Ψ_{11} (e^{- i ω}) |}^{2}}] = \log | 1 + \frac{{| Ψ_{12} (e^{- i ω}) |}^{2}}{{| Ψ_{11} (e^{- i ω}) |}^{2}} |

(5)

If $Y_{t}$ does not Granger-cause $X_{t}$ , then $M_{Y_{t} \to X_{t}} (ω) = 0$ . Referring to Breitung and Candelon (2006), equation (5) is equal to

Ψ_{12} (e^{- i ω}) = | θ_{12} (e^{- i ω}) | = | \sum_{k = 1}^{p} θ_{12, k} \cos (k ω) - \sum_{k = 1}^{p} θ_{12, k} \sin (k ω) | = 0

(6)

This suggests that $\sum_{k = 1}^{p} θ_{12, k} \cos (k ω) = 0$ and $\sum_{k = 1}^{p} θ_{12, k} \sin (k ω) = 0$ . To better examine the causal relationship between them, equation (6) can also be expressed as a linear restriction. We will conduct our research using the Wald statistic, the HJC information criterion and the bootstrap method. Further details can be found in Breitung and Candelon (2006) and Hatemi-J (2012). Therefore, their causality can be tested in time and frequency domains.

Data

We consider all news-based U.S. categorical economic policy uncertainty indices, namely, Economic Policy Uncertainty (EPU), taxes (Taxes), government spending (GS), health care (HC), national security (NS), entitlement programs (EP), regulation (Regulation), trade policy (TP), sovereign debt and currency crises (SD-CC). All monthly CEPU indicators, covering January 1986 to January 2022, can be obtained from the website of Economic Policy Uncertainty.³ The start date of the sample period depends on the availability and accessibility of data.

First, we plot the historical fluctuations in Figure 1. MPU is less volatile, with values below 500, while it shows some common spikes after major political and financial events. For example, during the 1991 Gulf War and the 2003 Gulf War, MPU, SD-CC, TP and NS fluctuated drastically. As tariff and trade policy tensions between the U.S. and China escalated, MPU, EP and TP fluctuated sharply. The descriptive statistics are presented in Table 1. All CEPUs had positive mean values and were right-skewed, indicating that they were distributed asymmetrically. Moreover, their kurtosis is greater than three, denoting leptokurtic distributions. The Jarque–Bera (JB) test shows that there is no indication of a Gaussian distribution for all the series data at the 1% significance level. The ADF test results show that all CEPUs are stationary, whereas the KPSS test results show that all CEPUs are non-stationary. However, given the model requirements, we took the first differences to make them stationary for the next analysis. We also calculated Spearman’s rank correlation coefficients which are shown in Table 2. Overall, EPU and taxes have the highest correlation coefficients, followed by taxes and GS. MPU has the highest correlation coefficient with EPU, followed by GS, and the lowest correlation coefficient with TP.

Figure 1.

MPU and all CEPUs.

Table 1.

Descriptive Statistics.

	Max.	Min.	Mean	SD.	Cov.	Skew.	Kurt.	JB	ADF	KPSS
MPU	407.9409	16.5745	91.7481	58.4113	0.6366	1.7583	7.4374	578.3508***	−5.0754***	0.1807**
EPU	503.0123	37.2660	104.2713	52.9561	0.5079	2.6328	15.2386	3202.6000***	−2.9427***	1.1475***
Taxes	471.8996	24.4412	109.7052	68.9159	0.6282	1.7503	6.9640	504.5839***	−3.5962***	0.7128***
GS	635.2718	5.7772	100.2021	93.8318	0.9364	2.4859	10.8427	1555.7000***	−5.6983***	0.9404***
HC	1030.7000	15.2818	133.6981	115.7120	0.8655	2.8591	16.2109	3738.7000***	−4.3921***	0.5740***
NS	758.2634	23.7371	94.7358	77.7861	0.8211	3.8353	24.3542	9288.6000***	−5.3775***	0.4225***
EP	1118.8000	14.3781	126.0768	123.1119	0.9765	3.5163	20.6600	6519.0000***	−5.2245***	0.4826***
Regulation	384.3900	28.1934	110.0440	56.0525	0.5094	1.5533	6.3128	372.1196***	−3.9530***	0.9948***
TP	1946.7000	7.6726	126.1144	183.9330	1.4585	4.5876	33.1732	17944.0000***	−6.6973***	1.9156***
SD-CC	1492.8000	0.0000	104.7833	185.9372	1.7745	4.2439	25.9695	10819.0000***	−8.0234***	0.3753***

Notes: *, ** and *** denote rejection of the null hypothesis at the 10%, 5% and 1% significance levels, respectively.

Table 2.

Spearman’s Rank Correlation Coefficients Among MPU and CEPU.

	MPU	EPU	Taxes	GS	HC	NS	EP	Regulation	TP	SD-CC
MPU	1	0.6760***	0.4307***	0.4441***	0.3038***	0.3684***	0.3323***	0.2684***	0.1600***	0.3013***
EPU		1	0.7500***	0.6287***	0.6240***	0.5919***	0.6104***	0.5702***	0.3181***	0.2473***
Taxes			1	0.7289***	0.7172***	0.4261***	0.7151***	0.4551***	0.2248***	0.1394**
GS				1	0.6377***	0.4237***	0.6615***	0.3604***	0.1305**	0.1514**
HC					1	0.3508***	0.7282***	0.3954***	0.1897***	0.0977*
NS						1	0.3160***	0.3393***	0.2287***	0.1179*
EP							1	0.4143***	0.1196*	0.1664*
Regulation								1	0.2262***	0.1664***
TP									1	0.1009*
SD-CC										1

Notes: *, ** and *** indicate statistical significance at the 10%, 5% and 1% levels, respectively.

Empirical Findings

Time-Domain Results

First, we applied the conventional Granger causality test to investigate the causal relationship between MPU and other CEPUs. The results are presented in Table 3. Panel A of Table 3 shows the effects of the other CEPUs on MPU. The null hypothesis of the Granger causality test from other CEPUs to MPU cannot be rejected at the 10% significance level, meaning that there is hardly any evidence of Granger causality from other CEPUs to MPU.

Table 3.

Time-Domain Results.

	Wald	1%	5%	10%
Panel A: Causality from CEPUs to MPU
EPU ↛ MPU	0.4068	9.6297	6.0448	4.6331
Taxes ↛ MPU	4.7705	13.0421	9.6741	7.9971
GS ↛ MPU	4.2498	15.2656	10.1900	7.8353
HC ↛ MPU	3.2896	15.0273	9.8132	7.7681
NS ↛ MPU	0.1936	7.3469	4.0913	2.7051
EP ↛ MPU	3.5008	15.3191	10.3080	8.0854
Regulation ↛ MPU	7.4767	12.4048	9.8210	8.0727
TP ↛ MPU	0.9012	7.1599	3.8800	2.5683
SD-CC ↛ MPU	5.2578	14.1432	9.6909	7.8467
Panel B: Causality from MPU to CEPUs
MPU ↛ EPU	2.0431	9.4624	6.2341	4.9018
MPU ↛ taxes	11.1693**	14.0986	9.7077	7.9397
MPU ↛ GS	10.9846**	13.3175	9.4489	7.6960
MPU ↛ HC	15.8310***	14.5425	9.6250	7.8416
MPU ↛ NS	9.0017***	8.1497	4.1661	2.8351
MPU ↛ EP	29.6586***	13.6367	9.1989	7.6486
MPU ↛ regulation	33.7454***	13.7287	9.4615	7.7540
MPU ↛ TP	0.0871	7.2165	3.9111	2.8560
MPU ↛ SD-CC	3.2559	15.5623	10.1808	8.1553

Notes: *, ** and*** denote statistical significance at the 10%, 5% and 1% levels, respectively.

However, Panel B of Table 3 provides evidence of significant causality from MPU to other CEPUs. Although we found that causality cannot be detected from MPU to EPU, TP and SD-CC, the null hypothesis of non-causality running from MPU to HC, NS, EP, and Regulation can be rejected at a significance level of 1%. In addition, at the 5% significance level, MPU had an impact on taxes and GS. Our results seem to be consistent with Thiem (2018), who employed the connectedness index of Diebold and Yilmaz (2012) and found strong connections between MPU and HC, NS and fiscal policy uncertainty. Our results are also supported by Jiang et al. (2019), who applied the TVP-VAR method and found fiscal (taxes or spending) policy uncertainty.

Frequency-Domain Results

To observe the long- and short-run effects, and response of MPU apropos other CEPUs, the frequency-based Granger causality test was employed for further analysis. We denote the frequency $ω$ on the horizontal axis by periodicity $T$ ( $T = \frac{2 π}{ω}, ω \in (0, π)$ ). According to Breitung and Candelon (2006), the result at $ω = 2.5$ usually indicates a short-term and temporary relation, while at $ω = 0.5$ implies a long-term and permanent impact. The result at $ω = 1.5$ implies a medium-term linkage.

First, in Figure 2, the null hypothesis of non-causality between MPU and EPU cannot be rejected for all the frequencies. One possible explanation is that EPU is a comprehensive indicator, and its effect and response to MPU may be neutralised by different CEPUs. As shown in Figure 3, the TP and MPU cannot impact each other. In addition, in Figure 4, we can barely find any evidence of bi-directional Granger causality between MPU and SD-CC. Eijffinger and Karataş (2012) and Erler et al. (2015) contended that different economies can adopt appropriate monetary policies to avoid possible output losses when currency crises occur. However, the uncertainty produced by adjusting monetary policies may be quite short-lived and thus have no significant short- and long-term impacts on SD-CC.