This article explores the extent to which economic policy uncertainty (EPU) influences selected macroeconomic variables in South Africa (SA).

To this end, I construct a constant parameter vector autoregressive (VAR) model and a time-varying parameter (TVP) VAR model, where the latter model evaluates if the impact of uncertainty on the macroeconomic variables has changed over time.

The models are estimated using quarterly South African data over the period 1990 to 2015, which include industrial production growth, consumer price inflation, 10-year government bond yield, real effective exchange rate, and economic policy uncertainty. Cholesky ordering of the variables are imposed to recover the orthogonal shocks.

The results of the constant parameter VAR model suggest that an unanticipated positive shock to the uncertainty index results in a decline in industrial production and real effective exchange rate, while fostering an increase in the general price level and 10-year government bond yield. Time-varying impulse responses show that the impact of uncertainty shocks on the selected macroeconomic variables has declined systematically over time. This is perhaps intuitive as the new unanticipated information is gradually picked up by media over time and incorporated into rational agents’ decision-making.

The transmission of a positive uncertainty shock to the real economy has time-varying implications.

Over the past few years, a growing number of official local and international publications have mentioned political uncertainty as a major factor influencing macroeconomic dynamics in South Africa (SA). For instance, the most recent International Monetary Fund (IMF) World Economic Outlook publication released in October 2017 stated that ‘in South Africa, growth is projected to remain subdued … as heightened political uncertainty saps consumer and business confidence’. Given this, my article will attempt to quantify the impact of economic policy uncertainty (EPU) on selected macroeconomic variables in SA by constructing a constant parameter vector autoregressive (VAR) model. In addition, I will attempt to examine if this impact changes over time, by constructing a time-varying parameter (TVP) VAR model.

My article contributes to the emerging market literature in two ways. Firstly, it extends the evidence that uncertainty shocks result in drops in real activity in developing countries. Secondly, it provides evidence on the time-varying impact of EPU shocks on the macroeconomic dynamics of an emerging market economy (SA in this case), by allowing for the variance of the structural shock, as well as the coefficients to vary over time. To my knowledge, the analysis of the time-varying impact of uncertainty, especially within a South African context, has not been examined in previous studies. Redl (

As such, my article aims to fill the gap in literature by estimating a constant parameter VAR model and a TVP VAR model for the South African economy, employing a measure of uncertainty constructed by Hlatshwayo and Saxegaard (

The impulse responses in the constant parameter VAR model suggest that an unanticipated positive shock to the uncertainty index is associated with a decline in industrial production and REER, while fostering an increase in the general price level and 10-year government bond yield. Markov Chain Monte Carlo (MCMC) methods in the context of a Bayesian inference are employed, following Primiceri (

The remainder of the article is organised as follows. In Section 2, I review the literature on uncertainty shocks. Section 3 describes the EPU index used and its construction. Section 4 presents the estimation procedure and empirical results of the constant parameter VAR model. Section 5 presents the estimation procedure and results of the TVP VAR model, and Section 6 concludes.

Theoretical literature highlights three broad reasons for why uncertainty matters, which include real options, risk aversion and growth options effect. Uncertainty influences growth via negative channels (real options and risk aversion) and positive channels (growth options) (Bloom

The idea of the real options effect is that firms face a number of investment projects that they may delay in the event of uncertainty. However, certain prerequisites are needed for this to have macroeconomic implications: firms must be subject to adjustment costs that cannot be easily reversible, must have the ability to wait in terms of bringing their product to the market, and finally firms’ investment actions today should influence the returns to actions taken tomorrow (for example, operate in an environment characterised by decreasing-returns-to-scale technology). When uncertainty is high, these effects can dampen the reallocation of resources across firms aimed at enhancing productivity, as productive firms expand less, while unproductive firms contract less as both are now cautious. According to Bloom, Baker and Davis (

Since risk averse investors require a higher return in the event of uncertainty, this means that greater uncertainty results in increasing risk premia, which raises the cost of borrowing and hence the probability of defaulting. The increase in the cost of borrowing has negative microeconomic and macroeconomic growth implications, due to the amplification of financial stress (Arellano, Bai & Kehoe

An increase in uncertainty can also be interpreted as positive on a macroeconomic level, although empirical literature has consistently proved otherwise. The growth options effect hinges on the idea that uncertainty can encourage investment to the extent that it increases the size of the potential return (by creating call options). The possible reason for why this effect is rarely observed in empirical studies is perhaps due to the ‘ambiguity aversion’ of agents. Such agents tend to assume the worst case scenario of the possible distributions they consider, thus an increase in uncertainty concerning future profits would lower the worst-case return on the investments (Bianchi, Ilut & Schneider

Since the global financial crisis of 2008, a growing amount of empirical literature has dealt with the measurement of uncertainty and how it affects macroeconomic variables. Bloom (

Baker, Bloom and Davis (

An index of economic uncertainty for SA was developed by Redl (

Baker, Bloom and Davis (

Hlatshwayo and Saxegaard (

The exploration of the time-varying nature of macroeconomic dynamics to uncertainty shocks have been quite limited, especially for emerging market economies. Alessandri and Mumtaz (

I make use of the ‘news chatter’ EPU index constructed by Hlatshwayo and Saxegaard (^{1}

Economic policy uncertainty index.

While global pressures loomed large during 2008–2009 due to the global financial crisis, domestic pressures were also present. Domestic pressures in 2008 included electricity shortages during the early part of the year, due to the near collapse of the national electricity grid, and the split of the African National Congress (ANC) into various parties which aroused concerns by many as to whether South Africa would adopt more populist policies. Uncertainty increased in 2012 due to domestic pressure emerging from the heavy protest- infested Marikana mining tragedy (worse strike action since post-apartheid era), and external pressure from the Eurozone crisis (sparked concerns related to central banks of developed economies raising interest rates above zero). Concerns loomed in South Africa during the 2014–2015 period following deepening drought conditions affecting food production and prices, and the double replacement of the Finance Minister in four days in late 2015. The figure also displays the recessionary periods of the South African economy, which do not necessarily coincide with spikes in uncertainty, suggesting that the measure picks up aspects of uncertainty beyond economic volatility.

In this section, I estimate a structural vector autoregressive (SVAR) model with constant parameters to analyse the extent to which uncertainty impacts upon selected macroeconomic variables. The benchmark model that is estimated is represented as:

_{t} is a vector of _{0} is an intercept term, _{1},_{s} are _{t} ~ N(^{2}^{3}

To recover orthogonal shocks, I use the Cholesky ordering of variables as above, ordering slower moving variables first before faster moving variables (Redl

Impulse responses of selected macroeconomic variables to a positive economic policy uncertainty shock: (a) response of ip to epu; (b) response of cpi to epu; (c) response of yield to epu; (d) response of reer to epu.

Contribution of economic policy uncertainty index to forecast error variance.

The impulse responses in

Variance decompositions in

In order to capture the possible time-varying nature of macroeconomic dynamics in the South African economy to shocks in uncertainty, I estimate a time-varying parameter VAR model with stochastic volatility.^{4}^{5}

The five-variable TVP VAR model that is estimated for the period 1990Q1–2015Q4 is specified by:

_{t}_{1t},_{st} are _{t} ~ _{t}_{t} is a _{t} i.e. _{t} represents a lower-triangular matrix with the diagonal elements equal to 1, and _{t} is a matrix with diagonal elements _{1t}, … . . _{kt}, and off-diagonal elements equal to 0. In this VAR specification, the coefficients _{1t},… … . …,_{st}, the parameters _{t} and _{t} are all time-varying. To model the process for these time-varying parameters (following Primiceri _{t} be a stacked row vector of _{1t},… … . …, _{st}, simultaneous relations _{t} and

_{β}, Σ_{h}_{a}_{t}, h_{t}, a_{t}

The model is specified with two lags as in the constant parameter VAR model, and the following priors (which provide reasonable identification) are assumed for the

^{6}

Sample autocorrelations (a–f), sample paths (g–l) and posterior densities (m–r).

Estimation results for selected parameters.

Parameter | Mean | Standard deviation | 95% interval | Convergence diagnostics | Inefficiency |
---|---|---|---|---|---|

0.0011 | 0.0000 | [0.0010, 0.0012] | 0.427 | 1.38 | |

0.0011 | 0.0000 | [0.0010, 0.0012] | 0.143 | 2.17 | |

0.0055 | 0.0016 | [0.0034, 0.0095] | 0.116 | 20.97 | |

0.0057 | 0.0017 | [0.0034, 0.0098] | 0.451 | 33.24 | |

0.1076 | 0.0456 | [0.0343, 0.2111] | 0.037 | 120.33 | |

0.0058 | 0.0018 | [0.0035, 0.0107] | 0.495 | 47.54 |

The mean of the sampled values in the two windows are then compared. However, there should be a sufficient number of iterations between the two windows to reasonably assume that the two means are approximately independent. This method then produces a CD statistic that is computed as the difference between the two means divided by the asymptotic standard error of the difference.^{7}

The ^{8}_{h1}, which nevertheless indicates an overall efficient sampling for the parameters and state variables. Even for the parameter _{h1}, the inefficiency factor is about 120, which implies that I obtain

This section presents the quantitative empirical results for the TVP VAR model analysing the time-varying structure of the macroeconomic dynamics to shocks to EPU within a South African context. The South African economy experienced several different periods over the sample period from 1990Q1 to 2015Q4, from the transition period to democracy to the global financial crisis. Employing the TVP VAR, I investigate the time-varying structure of South African macroeconomic dynamics to shocks to uncertainty, as follows.

Posterior estimates for stochastic volatility of the structural shock: (a) ip; (b) cpi; (c) yield; (d) reer; (e) epu.

Overall, stochastic volatility contributes to the VAR estimation, identifying the structural shock with the appropriate variance of the shock size. In this case, the time-invariant VAR model estimates would result in biases in the error covariance matrix and the autoregressive coefficients because of misspecification of the dynamics of the parameters.

The impulse response analyses for the times series in the TVP VAR model specified above are provided in

Impulse responses of selected macroeconomic variables to a positive economic policy uncertainty shock: (a) ip; (b) cpi; (c) yield; (d) reer.

Impulse responses of selected macroeconomic variables to a positive economic policy uncertainty shock: (a) _{epu} ↑ → ip_{epu} ↑ → cpi_{epu} ↑ → yield_{epu} ↑ → reer

Industrial production at the end of 2009 was supported by the collapse of the rand and commodity prices emanating from the global financial crisis. The response of inflation to uncertainty has time variation and the impact of the response is initially negative in 2004Q4 (subdued external demand), while demonstrating slight inflationary pressure in the comparative periods. Thereafter, inflationary impact turned negative during 2009, amid the subdued domestic and global demand inflicted by the financial crisis, and turned positive by varying magnitudes in the other periods investigated. A 10-year government bond yield does not exhibit much time variation in response to an uncertainty shock, with the magnitude of the positive response of the yield following an unanticipated shock to uncertainty being more or less the same for the comparative periods. A positive shock to uncertainty is associated with an initial decline in the REER in all the comparative periods. However, the magnitude of the response differs across the periods, with the 2009Q4 period of the financial crisis displaying a more pronounced decline. Thereafter, the REER increases later on in the comparative periods, displaying varying magnitudes, before moderating to equilibrium.

Overall, the impact of uncertainty shocks on industrial production, the REER and the price level has moderated slightly over time (with the timing of the change in the price level coinciding with the introduction of inflation targeting in the economy), while the impact on the bond yield has remained fairly stable.

Elasticity of selected macroeconomic variables following a 1% positive economic policy uncertainty shock.

Variable | ip (%) | cpi (%) | yield (%) | reer (%) |
---|---|---|---|---|

2 period ahead | −0.19 | 0.50 | 0.025 | 1.00 |

4 period ahead | −0.20 | 0.35 | −0.025 | 0.10 |

8 period ahead | −0.10 | 0.10 | −0.005 | 0.00 |

12 period ahead | −0.05 | 0.05 | 0.00 | 0.00 |

The elasticities, especially for the 4, 8 12 period ahead, suggest that the CPI index appears to be more sensitive to uncertainty shocks compared to the other macroeconomic variables analysed. The elasticities of the macroeconomic variables to the 1% unanticipated EPU shock diminishes by quarter 12. Overall,

The contribution of my article to emerging market empirical literature is twofold. Firstly, it extends the evidence that uncertainty shocks result in drops in real activity in developing countries, in this case the South African economy. Secondly, it provides evidence of the time-varying impact of EPU shocks on the macroeconomic dynamics of the South African economy, by allowing for both the coefficients and the variance of structural shock to vary over time.

The constant parameter VAR model shows that an unanticipated positive shock to the uncertainty index is associated with a decline in industrial production and REER, while fostering an increase in the general price level and 10-year government bond yield. The TVP VAR model reports posterior estimates for stochastic volatility of the structural shock, displaying the dynamics of volatility over time, which differ across variables. In this case, the estimates of the constant parameter VAR model would result in biases in the covariance matrix for the disturbances and at the same time in the autoregressive coefficients because of the misspecification of the dynamics of the parameters. The time-varying impulse responses for the specified dates (1994Q4, 1999Q4, 2004Q4 and 2009Q4) show that the impact of uncertainty shocks on industrial production, the REER and the price level has moderated slightly over time (with the timing of the change of the price level coinciding with the introduction of inflation targeting in the economy), while the impact on the bond yield has remained fairly stable. Meanwhile, the time-varying impulse responses and associated elasticities of the macroeconomic variables for 2, 4, 8 and 12 period ahead show a diminishing impact over time, following an unanticipated shock to the EPU index. This is perhaps intuitive, as the new unanticipated information is gradually picked up by media and incorporated into rational agents’ decision-making.

I would like to thank Professor Nicola Viegi for his supervisory support.

The author has declared that no competing interest exist.

The author contributed wholly to the acquisition of data, analysis and interpretation of findings.

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Ethical clearance was not required for the study.

Data sharing is not applicable to this article as no new data were created or analysed in this study.

The views and opinions expressed in this article are those of the authors and do not necessarily reflect the views of the University of Pretoria.

List of variables used in analysis.

Variable | Brief description | Source |
---|---|---|

Industrial production | An index measuring output of the industrial sector of the economy, specifically manufacturing and mining. Constructed based on the contribution of each sector to gross domestic product. | South African Reserve Bank (SARB) |

Consumer price inflation (CPI index) | The official measure of inflation in South Africa, which measures changes over time in the general level of prices of consumer goods and services. | Statistics South Africa (Stats SA) |

10-year government bond yield | The return on investment on the South African government’s debt obligations (bonds) or the interest rate the government pays to borrow money for different lengths of time (10 years in this case). | South African Reserve Bank (SARB) |

Real effective exchange rate (REER) | Reflects the weighted average of South Africa’s currency (the rand) relative to an index or basket of other major currencies, adjusted for the effects of inflation. | South African Reserve Bank (SARB) |

Economic policy uncertainty (EPU) index | ‘News chatter’ measure of uncertainty – count of the number of articles that match a certain search algorithms relating to words related to aggregate political and economic uncertainty. | Hlatshwayo and Saxegaard ( |

Augmented Dickey-Fuller unit root test on the variables analysed.

Variable | Level |
@TREND |
|||
---|---|---|---|---|---|

Status | Status | ||||

ip | −1.136033 | 0.6992 | Non-stationary | 0.0352 | Trend stationary (at 5 and 10% level) |

cpi | −3.128463 | 0.0276 | Non-Stationary (at 1% level) | 0.0007 | Trend stationary |

yield | −1.255227 | 0.6479 | Non-stationary | 0.0409 | Trend stationary (at 5 and 10% level) |

reer | −1.787163 | 0.3850 | Non-stationary | 0.0833 | Trend stationary (at 10% level) |

epu | −6.402749 | 0.0000 | Stationary | 0.2425 | Stationary in level |

Note: Test critical values: -3.4963 (1% level), -2.8903 (5% level), -2.5822 (10% level).

Lag length criterion test.

Lag | LogL | LR | FPE | AIC | SC | HQ |
---|---|---|---|---|---|---|

0 | −141.2112 | NA | 1.61e-05 | 3.150234 | 3.417354 | 3.258208 |

1 | 223.2431 | 675.7590 | 1.36e-08 | −3.921730 | −2.986812 |
−3.543821 |

2 | 253.3658 | 52.71476 |
1.23e-08 |
−4.028454 |
−2.425736 | −3.380609 |

3 | 271.0052 | 29.03151 | 1.46e-08 | −3.875108 | −1.604591 | −2.957328 |

4 | 293.8701 | 35.25010 | 1.56e-08 | −3.830627 | −0.892312 | −2.642912 |

5 | 308.0867 | 20.43639 | 2.02e-08 | −3.605973 | 0.000141 | −2.148323 |

6 | 321.0282 | 17.25529 | 2.74e-08 | −3.354754 | 0.919160 | −1.627168 |

7 | 341.5271 | 25.19652 | 3.26e-08 | −3.260980 | 1.680732 | −1.263459 |

8 | 367.8181 | 29.57738 | 3.52e-08 | −3.287876 | 2.321635 | −1.020420 |

Note: Three of the criteria (LR test statistic, FPE, AIC) of the lag length criteria test suggest inclusion of two lags in the model.

, Indicates lag order selected by the criterion.

LR, sequential modified LR test statistic (each test at 5% level); FPE, Final prediction error; AIC, Akaike information criterion; SC, Schwarz information criterion; HQ, Hannan-Quinn information criterion.

Vector autoregressive residual serial correlation LM tests.

Lags | LM-Stat | Probability |
---|---|---|

1 | 28.32829 | 0.2930 |

2 | 33.33416 | 0.1229 |

3 | 29.29917 | 0.2516 |

4 | 22.99270 | 0.5780 |

5 | 26.50909 | 0.3808 |

Note: Probabilities from chi-square with 25 degrees of freedom. The null hypothesis of no serial correlation at lag order 1 to 5 cannot be rejected at all conventional levels of significance, which implies that using a lag length of 2 is appropriate. Null hypothesis has no serial correlation at lag order h. Sample: 1990Q1 2015Q4. Included observations: 102.

LM-Stat, Lagrange multiplier statistic.

Vector autoregressive residual heteroskedasticity test: No cross terms.

Joint test |
||
---|---|---|

Chi-squared | Probability | |

344.2033 | 330 | 0.2840 |

Note: The null hypothesis that the residuals are homoskedastic cannot be rejected at all conventional levels of significance. Sample: 1990Q1 2015Q4. Included observations: 102

Vector autoregressive residual normality tests.

Component | Jarque-Bera | Probability | |
---|---|---|---|

1 | 144.6638 | 2 | 0.0000 |

2 | 1.273699 | 2 | 0.5290 |

3 | 23.46664 | 2 | 0.0000 |

4 | 24.61567 | 2 | 0.0000 |

5 | 29.99686 | 2 | 0.0000 |

Joint | 224.0166 | 10 | 0.0000 |

Note: The joint null hypothesis that the residuals are multivariate normal cannot be accepted at all conventional levels of significance; however, normality of residuals is not necessary when generating impulse response functions. Orthogonalisation: Cholesky (Lutkepohl). Null hypothesis: residuals are multivariate normal. Sample: 1990Q1 2015Q4. Included observations: 102

Plot of variables used in analysis: (a) ip; (b) cpi; (c) yield; (d) reer); (e) epu.

Stability test for constant parameter vector autoregressive model.

The EPU index constructed by Sandile Hlatshwayo can be obtained from her website at

See

See

Primiceri (

See Nakajima (

MCMC algorithm sampling the posterior distribution π(β, _{β}_{a}_{h}

CD statistic computed as _{0} and _{1} represent the first and last draws respectively, _{o} = 1, _{1} = 5001, _{0} = 1000, _{1} = 5000,

The inefficiency factor is used to judge how well the MCMC chain mixes and is computed as _{k}_{m}