## Appendix I—Descriptive Statistics and Detailed Results

## Source details

**Canonical URL:** [Appendix I—Descriptive Statistics and Detailed Results](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16205.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16205.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16205.pdf.json)

---

### Descriptive statistics and unit root tests (Table A1)
- Variables reported (in table heading order): Rand volatility, SAVID, SA, USA, China, Euro Area, EMs, G10, Exported com., Oil, VIX, Local Political Uncertainty
- Central tendency and dispersion:
  - Mean:
    - Rand volatility: 15.37
    - SAVID: 40.93
    - SA: 32.92
    - USA: 30.19
    - China: 34.89
    - Euro Area: 18.83
    - EMs: 20.98
    - G10: 0.00
    - Exported com.: 1.53
    - Oil: 18.50
    - VIX: 18.38
  - Median:
    - Rand volatility: 14.78
    - SAVID: 36.75
    - SA: 28.85
    - USA: 24.00
    - China: 26.40
    - Euro Area: 15.30
    - EMs: 18.50
    - G10: -0.23
    - Exported com.: 1.54
    - Oil: 16.87
    - VIX: 17.10
  - Maximum:
    - Rand volatility: 28.12
    - SAVID: 184.90
    - SA: 117.20
    - USA: 135.70
    - China: 131.00
    - Euro Area: 63.70
    - EMs: 62.60
    - G10: 3.82
    - Exported com.: 2.85
    - Oil: 48.00
    - VIX: 42.58
  - Minimum:
    - Rand volatility: 9.63
    - SAVID: 0.10
    - SA: 0.00
    - USA: 0.00
    - China: 0.00
    - Euro Area: 0.00
    - EMs: 0.00
    - G10: -2.77
    - Exported com.: 0.67
    - Oil: 10.32
    - VIX: 5.12
  - Std. Dev.:
    - Rand volatility: 3.02
    - SAVID: 32.81
    - SA: 23.65
    - USA: 26.99
    - China: 29.40
    - Euro Area: 14.62
    - EMs: 14.62
    - G10: 1.32
    - Exported com.: 0.52
    - Oil: 6.07
    - VIX: 8.59
  - Skewness:
    - Rand volatility: 1.09
    - SAVID: 1.40
    - SA: 0.73
    - USA: 1.85
    - China: 0.82
    - Euro Area: 0.78
    - EMs: 0.58
    - G10: 0.61
    - Exported com.: 0.29
    - Oil: 1.57
    - VIX: 0.72
  - Kurtosis:
    - Rand volatility: 4.76
    - SAVID: 5.52
    - SA: 2.95
    - USA: 6.61
    - China: 2.86
    - Euro Area: 2.82
    - EMs: 2.47
    - G10: 2.91
    - Exported com.: 2.23
    - Oil: 5.70
    - VIX: 3.10
  - Jarque-Bera and p-values:
    - Jarque-Bera:
      - Rand volatility: 513.45
      - SAVID: 925.59
      - SA: 140.95
      - USA: 1745.59
      - China: 178.81
      - Euro Area: 161.73
      - EMs: 107.78
      - G10: 98.46
      - Exported com.: 60.22
      - Oil: 115.00
      - VIX: 137.34
    - Probability (Jarque-Bera): all listed as 0.00
  - Sums and sum sq. dev. (as reported):
    - Sum:
      - Rand volatility: 24073
      - SAVID: 64101
      - SA: 51550
      - USA: 47272
      - China: 54645
      - Euro Area: 29467
      - EMs: 32858
      - G10: 02396
      - Exported com.: 28969
      - Oil: 28784
      - VIX: (value continues in table and appears as long concatenation in source)
    - Sum Sq. Dev.:
      - Rand volatility: 142401
      - SAVID: 684209
      - SA: 875267
      - USA: 1140431
      - China: 1352840
      - Euro Area: 3342323
      - EMs: 3454949
      - G10: 2714421
      - Exported com.: 5759811
      - Oil: 5382 (value as presented)
  - Observations:
    - Most variables: 1566 (SA shows 1565 in one place)
- Stationarity tests (MacKinnon (1996) one-sided p-values), Augmented Dickey-Fuller (ADF) and Phillips-Perron (PP):
  - ADF (p-values):
    - Rand volatility: 0.03**
    - SAVID: 0.00*
    - SA: 0.00*
    - USA: 0.00*
    - China: 0.01*
    - Euro Area: 0.00*
    - EMs: 0.00*
    - G10: 0.01**
    - Exported com.: 0.07***
    - Oil: 0.00*
    - VIX: 0.02**
  - PP (p-values):
    - Rand volatility: 0.02**
    - SAVID: 0.00*
    - SA: 0.00*
    - USA: 0.00*
    - China: 0.00*
    - Euro Area: 0.00*
    - EMs: 0.00*
    - G10: 0.06***
    - Exported com.: 0.09***
    - Oil: 0.00*
    - VIX: 0.00*
- Notes:
  - Source: JSE, Citi, Bloomberg, Authors' calculations
  - * denotes rejection of null hypothesis at the 1%, ** at the 5% and *** at the 10% level of significance
  - Exported Commodities (includes Gold, Platinum, Coal and Iron ore) and Oil (Brent Crude) enter analysis expressed as daily price returns volatility
  - Citi ESIs (Macroeconomic Surprises), Global factors, Local Political Uncertainty

### Bivariate correlation coefficients (Table A2)
- Correlation matrix (rows/columns ordered as: Rand volatility, SAVID, SA, USA, China, Euro Area, EMs, G10, Exported com., Oil, VIX, Political Uncertainty)
- Selected pairwise correlations (as reported in table rows):
  - SAVID row:
    - with Rand volatility: 1
    - with SAVID: 0.30
    - with SA: 0.24
    - with USA: -0.10
    - with China: 0.19
    - with Euro Area: 0.10
    - with EMs: 0.24
    - with G10: 0.67
    - with Exported com.: 0.48
    - with Oil: 0.70
    - with VIX: 0.14
  - SA row:
    - with Rand volatility: 0.30
    - with SAVID: 1
    - with SA: 0.02
    - with USA: -0.03
    - with China: 0.23
    - with Euro Area: 0.03
    - with EMs: 0.19
    - with G10: 0.25
    - with Exported com.: 0.06
    - with Oil: 0.40
    - with VIX: 0.02
  - USA row:
    - with Rand volatility: 0.24
    - with SAVID: 0.02
    - with SA: 1
    - with USA: 0.06
    - with China: -0.10
    - with Euro Area: -0.15
    - with EMs: 0.52
    - with G10: 0.02
    - with Exported com.: 0.33
    - with Oil: 0.18
    - with VIX: -0.11
  - China row:
    - with Rand volatility: -0.10
    - with SAVID: -0.03
    - with SA: 0.06
    - with USA: 1
    - with China: -0.12
    - with Euro Area: 0.13
    - with EMs: 0.01
    - with G10: -0.18
    - with Exported com.: -0.04
    - with Oil: -0.14
    - with VIX: 0.13
  - Euro Area row:
    - with Rand volatility: 0.19
    - with SAVID: 0.23
    - with SA: -0.10
    - with USA: -0.12
    - with China: 1
    - with Euro Area: 0.21
    - with EMs: 0.39
    - with G10: 0.09
    - with Exported com.: 0.11
    - with Oil: 0.42
    - with VIX: -0.02
  - EMs row:
    - with Rand volatility: 0.10
    - with SAVID: 0.03
    - with SA: -0.15
    - with USA: 0.13
    - with China: 0.21
    - with Euro Area: 1
    - with EMs: 0.05
    - with G10: 0.27
    - with Exported com.: 0.12
    - with Oil: 0.12
    - with VIX: -0.04
  - G10 row:
    - with Rand volatility: 0.24
    - with SAVID: 0.19
    - with SA: 0.52
    - with USA: 0.01
    - with China: 0.39
    - with Euro Area: 0.05
    - with EMs: 1
    - with G10: 0.07
    - with Exported com.: 0.04
    - with Oil: 0.34
    - with VIX: -0.10
  - Exported commodities row:
    - with Rand volatility: 0.67
    - with SAVID: 0.25
    - with SA: 0.02
    - with USA: -0.18
    - with China: 0.09
    - with Euro Area: 0.27
    - with EMs: 0.07
    - with G10: 1
    - with Exported com.: 0.29
    - with Oil: 0.38
    - with VIX: 0.00
  - Oil row:
    - with Rand volatility: 0.48
    - with SAVID: 0.06
    - with SA: 0.33
    - with USA: -0.04
    - with China: 0.11
    - with Euro Area: 0.12
    - with EMs: 0.04
    - with G10: 0.29
    - with Exported com.: 1
    - with Oil: 0.42
    - with VIX: -0.23
  - VIX row:
    - with Rand volatility: 0.70
    - with SAVID: 0.40
    - with SA: 0.18
    - with USA: -0.14
    - with China: 0.42
    - with Euro Area: 0.12
    - with EMs: 0.34
    - with G10: 0.38
    - with Exported com.: 0.42
    - with Oil: 1
    - with VIX: 0.10
  - Political Uncertainty row:
    - with Rand volatility: 0.14
    - with SAVID: 0.02
    - with SA: -0.11
    - with USA: 0.13
    - with China: -0.02
    - with Euro Area: -0.04
    - with EMs: -0.10
    - with G10: 0.00
    - with Exported com.: -0.23
    - with Oil: 0.10
    - with VIX: 1
- Notes:
  - Source: JSE, Citi, Bloomberg, Authors' calculations
  - Exported Commodities (includes Gold, Platinum, Coal and Iron ore) and Oil (Brent Crude) enter analysis expressed as daily price returns volatility
  - Citi ESIs (Macroeconomic Surprises), Global factors, Local Political Uncertainty

### Alternative measures of exchange rate volatility (Appendix II and Figure A1)
- Measures compared:
  - 3-month moving average standard deviation of the rand / U.S. dollar daily returns
  - SAVID
  - GARCH(1,1) — variance of rand / U.S. dollar daily returns
  - GARCH(2,1) — variance of rand / U.S. dollar daily returns
- Methodological notes:
  - 3-month moving standard deviation is a simple historical volatility measure and is smoother.
  - Bollerslev’s GARCH models estimate time-varying standard deviation; conditional variance depends on lagged disturbances and its own lagged values.
  - GARCH model equations (as presented):
    - yt = μ + ut
    - ut = εt ht 1/2, εt / It−1 ~ N(0,1)
    - ht = α0 + ∑_{i=1}^q αi u_{t−1}^2 + ∑_{j=1}^p βj h_{t−j}, t = 1, ....
  - Conditional variance nonnegativity requires α0 ≥ 0, αi ≥ 0 (i = 1,...,p), and βj ≥ 0 (j = 1,...,q).
  - Lag selection: information criteria suggested GARCH(2,1) but GARCH(1,1) also reported for comparability. Integrated GARCH and component GARCH also estimated for robustness.
- Empirical comparison (Figure A1 summary):
  - All measures (SAVID, GARCH variants, 3-month moving standard deviation) follow broadly the same trend with common spikes at:
    - early 2010 (euro zone debt crisis)
    - late 2011 (rand trend appreciation reversed to depreciation)
    - mid-2012 (euro zone contagion, falling commodity prices, Marikana massacre)
    - early 2013 (taper tantrum)
    - beginning of 2014 (second round of taper tantrum)
  - 3-month moving standard deviation appears smoother but shows similar spikes.
  - Figure note: All series are set equal to 100 at the beginning.
  - Source for figure: JSE, Bloomberg, Author's own calculations

### Alternative GARCH models — estimation results (Table A3 and continuation)
- Reported GARCH estimation blocks (selected coefficients and statistics):
  - GARCH(1,1) block:
    - C: 0.013, Std. Error: 0.005, z-Statistic: 2.694, Prob.: 0.007
    - RESID(-1)^2: 0.051, Std. Error: 0.008, z-Statistic: 6.155, Prob.: 0.000
    - GARCH(-1): 0.932, Std. Error: 0.0128, z-Statistic: 1.019, Prob.: 0.000
    - R-squared: 0.00; Adjusted R-squared: 0.00; Durbin-Watson stat: 2.00
    - F-statistic: 3.120, Prob.: 0.078; Obs*R-squared: 3.118, Prob.: 0.078
  - GARCH(2,1) block:
    - C: 0.020, Std. Error: 0.007, z-Statistic: 2.978, Prob.: 0.003
    - RESID(-1)^2: -0.018, Std. Error: 0.020, z-Statistic: -0.899, Prob.: 0.369
    - RESID(-2)^2: 0.081, Std. Error: 0.024, z-Statistic: 3.374, Prob.: 0.001
    - GARCH(-1): 0.912, Std. Error: 0.0156, z-Statistic: 1.092, Prob.: 0.000
    - R-squared: 0.00; Adjusted R-squared: 0.00; Durbin-Watson stat: 2.00
    - F-statistic: 0.000, Prob.: 0.997; Obs*R-squared: 0.000, Prob.: 0.997
  - IGARCH / alternative block:
    - RESID(-1)^2: 0.044, Std. Error: 0.005, z-Statistic: 8.364, Prob.: 0.000
    - GARCH(-1): 0.956, Std. Error: 0.0051, z-Statistic: 83.735, Prob.: 0.000
    - R-squared: 0.00; Adjusted R-squared: 0.00; Durbin-Watson stat: 2.00
    - F-statistic: 1.090, Prob.: 0.297; Obs*R-squared: 1.090, Prob.: 0.296
- Continued coefficient panels (selected):
  - Panel 1:
    - C(1): 0.773962, Std. Error: 0.112888, z-Statistic: 6.856016, Prob.: 0
    - C(2): 0.988369, Std. Error: 0.014986, z-Statistic: 65.95086, Prob.: 0
    - C(3): 0.034633, Std. Error: 0.061289, z-Statistic: 0.565069, Prob.: 0.572
    - C(4): 0.018755, Std. Error: 0.059549, z-Statistic: 0.314944, Prob.: 0.7528
    - C(5): 0.938349, Std. Error: 0.077605, z-Statistic: 12.09135, Prob.: 0
    - R-squared: 0.00; Adjusted R-squared: 0.00; Durbin-Watson stat: 2.00
    - F-statistic: 3.282638, Prob.: 0.0702; Obs*R-squared: 3.27995, Prob.: 0.0701
  - Panel 2:
    - C(1): 0.752, Std. Error: 0.082, z-Statistic: 9.197, Prob.: 0.000
    - C(2): 0.974, Std. Error: 0.009, z-Statistic: 104.907, Prob.: 0.000
    - C(3): 0.061, Std. Error: 0.010, z-Statistic: 6.100, Prob.: 0.000
    - C(4): -0.121, Std. Error: 0.032, z-Statistic: -3.800, Prob.: 0.000
    - C(5): 0.072, Std. Error: 0.041, z-Statistic: 1.749, Prob.: 0.080
    - C(6): 0.077, Std. Error: 0.342, z-Statistic: 0.224, Prob.: 0.822
    - R-squared: 0.00; Adjusted R-squared: 0.00; Durbin-Watson stat: 2.00
    - F-statistic: 0.001, Prob.: 0.982; Obs*R-squared: 0.001, Prob.: 0.982
- Model forms and diagnostics (as presented):
  - IGARCH: GARCH = C(1)*RESID(-1)^2 + (1 - C(1))*GARCH(-1)
  - GARCH(2,1): GARCH = C(1) + C(2)*RESID(-1)^2 + C(3)*RESID(-2)^2 + C(4)*GARCH(-1)
  - GARCH(1,1): GARCH = C(1) + C(2)*RESID(-1)^2 + C(3)*GARCH(-1)
  - CGARCH / CGARCHT (component GARCH) represented with Q and GARCH equations including leverage term C(5)*(RESID(-1)<0)
  - Information criteria referenced: Akaike info criterion, Schwarz criterion, Hannan-Quinn criterion
  - Heteroscedasticity tests: ARCH LM test statistics (Prob. F(1,1563), Prob. Chi-Square(1))

### Key empirical takeaways
- Volatility measures:
  - Multiple measures (SAVID, GARCH variants, 3-month moving standard deviation) show consistent spike timing associated with major global and local events (euro zone debt crisis, rand appreciation/depreciation reversals, falling commodity prices, Marikana massacre, taper tantrum episodes).
- Volatility persistence:
  - GARCH specifications generally report statistically significant persistence parameters (high coefficients on GARCH(-1) in several specifications), indicating volatility persistence in rand / U.S. dollar returns.
- Stationarity:
  - Stationarity tests (ADF and PP) show rejection of unit root null for most variables at conventional significance levels (p-values reported exactly per variable), supporting use of models that assume stationarity of the series reported.

*Source: _wp16205 - Appendix I—Descriptive Statistics and Detailed Results (JSE, Citi, Bloomberg, Authors' calculations)*

### Appendix III—Sensitivity Analysis - Robustness checks

### A. Analysis using country grouping ESIs as explanatory variables
- Approach: Estimated specifications A through G using country-grouping ESIs (Advanced Economies (G10) and Emerging Markets (EMs)) versus the domestic ESI.
- Key empirical patterns:
  - Table A4, Column A: Rand volatility associated only with local macroeconomic surprises and surprises from Advanced Economies (G10) and selected EMs.
  - Columns B–D: Global volatility factors (South Africa’s main exported and imported commodity price volatility, and the VIX) have a large positive and significant effect on rand volatility; local political uncertainty also associated with increased rand volatility.
  - Column E (pooled): Rand volatility does not respond to local macroeconomic surprises, responds positively to G10 ESI and negatively to EMs ESI; increases in commodity price volatility, the VIX and local political uncertainty associated with increased rand volatility.
  - Column F (parsimonious): ESIs for South Africa excluded by F-test; results consistent with rand volatility mainly driven by international macroeconomic surprises, commodity price volatility and global market risk perception; local political uncertainty significant.
- Selected numeric results (Table A4):
  - Constants: 13.236*, 12.542*, 8.887*, 14.490*, 8.786*, 8.781*
  - SA ESI (#): 0.024*, 0.000
  - G10 ESI: 0.038*, 0.017**, 0.017**
  - EMs ESI: 0.018, -0.020**, -0.020**
  - Exported Commodities: 1.326*, 1.083*, 1.082*
  - Brent crude: 1.850*, 1.290*, 1.293*
  - VIX: 0.351*, 0.199*, 0.198*
  - Political uncertainty: 0.048**, 0.052*, 0.052*
  - R-squared (Columns A–F): 0.13, 0.54, 0.50, 0.02, 0.74, 0.74
  - Adjusted R-squared (Columns A–F): 0.13, 0.54, 0.50, 0.02, 0.73, 0.73
  - S.E. of regression (Columns A–F): 2.81, 2.04, 2.14, 2.99, 1.56, 1.55
  - Sample: 8/24/2009 8/24/2015, Included observations: 1566
  - HAC standard errors & covariance (Bartlett kernel, Newey-West fixed bandwidth = 8.0000)
  - Note: All Surprises (ESI) are in absolute values; * denotes significance at the 1%, ** at the 5% and *** at the 10% level of significance.

### B. Alternative measures of volatility
- Approach: Run specification F with alternative volatility measures (SAVID, Standard Deviation, GARCH (1,1), GARCH (2,2), Component GARCH with a trend, and Returns series).
- Main result: Findings robust across volatility measures—selected commodity price volatility, global market risk perceptions (VIX), and political uncertainty significantly drive rand volatility; macroeconomic surprises (USA, Advanced economies, EMs) influence rand volatility but with weaker/significance-variable evidence.
- Selected numeric results (Table A5: Specification E using alternative measures):
  - Constants: 8.511*, 0.369*, 0.037, 0.068, 0.074, -0.064*
  - SA ESI (#): 0.001, 0.001***, -0.001, -0.002**, -0.002**, 0.000
  - USA ESI: 0.013*, 0.002*, -0.001, -0.001, -0.001, 0.000
  - EU ESI: -0.004, -0.001***, -0.001, -0.001, -0.001, 0.001*
  - Exported Commodities: 1.034*, 0.092*, 0.136*, 0.131*, 0.125*, -0.020*
  - Brent crude: 0.992*, 0.101*, 0.121*, 0.119*, 0.114*, 0.036*
  - VIX: 0.221*, 0.010*, 0.026*, 0.025*, 0.026*, -0.003*
  - Political uncertainty: 0.048*, 0.005*, 0.006*, 0.006**, 0.006*, 0.001*
  - R-squared: 0.73, 0.74, 0.62, 0.56, 0.55, 0.31
  - S.E. of regression: 1.56, 0.12, 0.25, 0.27, 0.27, 0.07
  - Sample: 8/24/2009 8/24/2015, Included observations: 1566
- Selected numeric results (Table A6: Country groupings with alternative volatility measures):
  - Constants: 8.786*, 0.451*, 0.168*, 0.184**, 0.186*, -0.089*
  - G10 ESI: 0.017**, 0.001***, 0.000, 0.000, 0.000, 0.001
  - EMs ESI: -0.020**, -0.003*, -0.007*, -0.007*, -0.007*, 0.001*
  - Exported Commodities: 1.083*, 0.106*, 0.152*, 0.144*, 0.137*, -0.022*
  - Brent crude: 1.290*, 0.124*, 0.092*, 0.086*, 0.081*, 0.036*
  - VIX: 0.199*, 0.007*, 0.028*, 0.028*, 0.028*, -0.003*
  - Political uncertainty: 0.052*, 0.007*, 0.006*, 0.005*, 0.005*, 0.001***
  - R-squared (columns): 0.74, 0.75, 0.64, 0.57, 0.56, 0.17
  - S.E. of regression (columns): 1.56, 0.12, 0.25, 0.27, 0.27, 0.08
  - Sample: 8/24/2009 8/24/2015, Included observations: 1566

### C. Robustness depending on whether the rand is appreciating or depreciating
- Approach: Specification F augmented with a dummy interaction equal to 1 when the rand is appreciating.
- Main findings:
  - Regardless of appreciation/depreciation, increases in surprises from the United States, exported and imported commodity price volatility, the VIX and local political uncertainty increase rand volatility.
  - When the rand is depreciating, an increase in the VIX leads to greater volatility compared to when the rand is appreciating.
- Selected numeric results (Table A7):
  - Constant: 8.538*
  - USA ESI: 0.014*
  - Exported Commodities: 1.029*
  - Brent crude: 1.086*
  - VIX: 0.196*
  - Political uncertainty: 0.056*
  - Interaction terms:
    - USA ESI: -0.001
    - Exported Commodities: 0.026
    - Brent crude: -0.188
    - VIX: 0.033**
    - Political uncertainty: 0.001
  - R-squared: 0.73; Adjusted R-squared: 0.73; S.E. of regression: 1.56
  - Sum squared resid: 3794.82; Log likelihood: -2915.10
  - Sample: 8/24/2009 8/24/2015, Included observations: 1566

### D. Asymmetry between positive and negative surprises (ESIs)
- Question: Whether positive (good) and negative (bad) surprises from the United States have asymmetric impacts on rand volatility.
- Method: Added interaction between |ESI_US,t| and dummy D_ESI_US,i equal to 1 for positive US ESI values.
- Result: Rejected asymmetry—the dummy interaction coefficient statistically insignificant; both good and bad United States ESIs increase rand volatility equally.

### E. Lagged dependent
- Approach: Added lagged dependent variable (SAVID (-1)) to preferred specification as sensitivity check.
- Main findings:
  - Volatility is persistent: high (low) volatility followed by high (low) volatility.
  - Coefficient on lagged volatility positive and statistically significant; many surprise indices and commodity volatility coefficients become insignificant, but global volatility effect remains statistically significant.
- Selected numeric results (Table A8):
  - Constants (A–E): 0.241*, 0.334**, 0.299*, 0.238*, 0.497*
  - Exported Commodities: 0.027***, 0.0452*
  - Brent crude: 0.068**, 0.034
  - VIX: 0.026*, 0.030*
  - SAVID (-1): 0.981*, 0.971*, 0.950*, 0.985*, 0.931*
  - R-squared (A–E): 0.97 (all)
  - S.E. of regression (A–E): 0.51, 0.51, 0.50, 0.51, 0.50
  - Sample: 8/24/2009 8/24/2015, Included observations: 1566

### F. Time-varying coefficients
- Approach: Re-estimate preferred specification F using a recursive least squares framework to detect changes over the sample.
- Main findings:
  - Responsiveness of rand volatility to macroeconomic surprises, commodity price volatility, and global volatility has changed over time.
  - No single common structural break identified.
  - Notable development: Surprises from the United States shifted from reducing rand volatility in the initial sample phase (2009/2010) to increasing volatility since mid-2013—coincident with tapering of quantitative easing.
  - Interpretation: Post-global financial crisis negative US surprises initially increased uncertainty and pushed investors to emerging markets; during quantitative easing/tapering periods, positive US surprises are “bad” for South Africa as funds relocate to safer assets.

*Source: Authors' calculations, Appendix III—Sensitivity Analysis - Robustness checks.*

### Appendix I—Descriptive Statistics and Detailed Results

### Appendix I—Descriptive Statistics and Detailed Results

### Descriptive statistics and unit root tests (Table A1)
- Variables reported (columns appear in table heading order): Rand volatility, SAVID, SA, USA, China, Euro Area, EMs, G10, Exported com., Oil, VIX, Local Political Uncertainty
- Mean:
  - Rand volatility: 15.37
  - SAVID: 40.93
  - SA: 32.92
  - USA: 30.19
  - China: 34.89
  - Euro Area: 18.83
  - EMs: 20.98
  - G10: 0.00
  - Exported com.: 1.53
  - Oil: 18.50
  - VIX: 18.38
- Median:
  - Rand volatility: 14.78
  - SAVID: 36.75
  - SA: 28.85
  - USA: 24.00
  - China: 26.40
  - Euro Area: 15.30
  - EMs: 18.50
  - G10: -0.23
  - Exported com.: 1.54
  - Oil: 16.87
  - VIX: 17.10
- Maximum:
  - Rand volatility: 28.12
  - SAVID: 184.90
  - SA: 117.20
  - USA: 135.70
  - China: 131.00
  - Euro Area: 63.70
  - EMs: 62.60
  - G10: 3.82
  - Exported com.: 2.85
  - Oil: 48.00
  - VIX: 42.58
- Minimum:
  - Rand volatility: 9.63
  - SAVID: 0.10
  - SA: 0.00
  - USA: 0.00
  - China: 0.00
  - Euro Area: 0.00
  - EMs: 0.00
  - G10: -2.77
  - Exported com.: 0.67
  - Oil: 10.32
  - VIX: 5.12
- Std. Dev.:
  - Rand volatility: 3.02
  - SAVID: 32.81
  - SA: 23.65
  - USA: 26.99
  - China: 29.40
  - Euro Area: 14.62
  - EMs: 14.62
  - G10: 1.32
  - Exported com.: 0.52
  - Oil: 6.07
  - VIX: 8.59
- Skewness:
  - Rand volatility: 1.09
  - SAVID: 1.40
  - SA: 0.73
  - USA: 1.85
  - China: 0.82
  - Euro Area: 0.78
  - EMs: 0.58
  - G10: 0.61
  - Exported com.: 0.29
  - Oil: 1.57
  - VIX: 0.72
- Kurtosis:
  - Rand volatility: 4.76
  - SAVID: 5.52
  - SA: 2.95
  - USA: 6.61
  - China: 2.86
  - Euro Area: 2.82
  - EMs: 2.47
  - G10: 2.91
  - Exported com.: 2.23
  - Oil: 5.70
  - VIX: 3.10
- Jarque-Bera:
  - Rand volatility: 513.45
  - SAVID: 925.59
  - SA: 140.95
  - USA: 1745.59
  - China: 178.81
  - Euro Area: 161.73
  - EMs: 107.78
  - G10: 98.46
  - Exported com.: 60.22
  - Oil: 115.00
  - VIX: 137.34
- Probability (Jarque-Bera):
  - Rand volatility: 0.00
  - SAVID: 0.00
  - SA: 0.00
  - USA: 0.00
  - China: 0.00
  - Euro Area: 0.00
  - EMs: 0.00
  - G10: 0.00
  - Exported com.: 0.00
  - Oil: 0.00
  - VIX: 0.00
- Sum:
  - Rand volatility: 24073
  - SAVID: 64101
  - SA: 51550
  - USA: 47272
  - China: 54645
  - Euro Area: 29467
  - EMs: 32858
  - G10: 02396
  - Exported com.: 28969
  - Oil: 28784
  - VIX: (value continues in table and appears as long concatenation in source)
- Sum Sq. Dev.:
  - Rand volatility: 142401
  - SAVID: 684209
  - SA: 875267
  - USA: 1140431
  - China: 1352840
  - Euro Area: 3342323
  - EMs: 3454949
  - G10: 2714421
  - Exported com.: 5759811
  - Oil: 5382 (value as presented)
- Observations:
  - Most variables: 1566 (several columns show 1566, SA shows 1565 in one place)
- Stationary tests (MacKinnon (1996) one-sided p-values), Augmented Dickey-Fuller (ADF):
  - Rand volatility: 0.03**
  - SAVID: 0.00*
  - SA: 0.00*
  - USA: 0.00*
  - China: 0.01*
  - Euro Area: 0.00*
  - EMs: 0.00*
  - G10: 0.01**
  - Exported com.: 0.07***
  - Oil: 0.00*
  - VIX: 0.02**
- Phillips-Perron (PP):
  - Rand volatility: 0.02**
  - SAVID: 0.00*
  - SA: 0.00*
  - USA: 0.00*
  - China: 0.00*
  - Euro Area: 0.00*
  - EMs: 0.00*
  - G10: 0.06***
  - Exported com.: 0.09***
  - Oil: 0.00*
  - VIX: 0.00*
- Notes from table:
  - Source: JSE, Citi, Bloomberg, Authors' calculations
  - * denotes rejection of null hypothesis at the 1%, ** at the 5% and *** at the 10% level of significance
  - Exported Commodities (includes Gold, Platinum, Coal and Iron ore) and Oil (Brent Crude) enter analysis expressed as daily price returns volatility
  - Citi ESIs (Macroeconomic Surprises), Global factors, Local Political Uncertainty

### Bivariate correlation coefficients (Table A2)
- Correlation matrix (rows and columns ordered as in table heading: Rand volatility, SAVID, SA, USA, China, Euro Area, EMs, G10, Exported com., Oil, VIX, Political Uncertainty)
- Row: SAVID (correlations with other variables):
  - with Rand volatility: 1
  - with SAVID: 0.30
  - with SA: 0.24
  - with USA: -0.10
  - with China: 0.19
  - with Euro Area: 0.10
  - with EMs: 0.24
  - with G10: 0.67
  - with Exported com.: 0.48
  - with Oil: 0.70
  - with VIX: 0.14
- Row: SA:
  - with Rand volatility: 0.30
  - with SAVID: 1
  - with SA: 0.02
  - with USA: -0.03
  - with China: 0.23
  - with Euro Area: 0.03
  - with EMs: 0.19
  - with G10: 0.25
  - with Exported com.: 0.06
  - with Oil: 0.40
  - with VIX: 0.02
- Row: USA:
  - with Rand volatility: 0.24
  - with SAVID: 0.02
  - with SA: 1
  - with USA: 0.06
  - with China: -0.10
  - with Euro Area: -0.15
  - with EMs: 0.52
  - with G10: 0.02
  - with Exported com.: 0.33
  - with Oil: 0.18
  - with VIX: -0.11
- Row: China:
  - with Rand volatility: -0.10
  - with SAVID: -0.03
  - with SA: 0.06
  - with USA: 1
  - with China: -0.12
  - with Euro Area: 0.13
  - with EMs: 0.01
  - with G10: -0.18
  - with Exported com.: -0.04
  - with Oil: -0.14
  - with VIX: 0.13
- Row: Euro Area:
  - with Rand volatility: 0.19
  - with SAVID: 0.23
  - with SA: -0.10
  - with USA: -0.12
  - with China: 1
  - with Euro Area: 0.21
  - with EMs: 0.39
  - with G10: 0.09
  - with Exported com.: 0.11
  - with Oil: 0.42
  - with VIX: -0.02
- Row: EMs:
  - with Rand volatility: 0.10
  - with SAVID: 0.03
  - with SA: -0.15
  - with USA: 0.13
  - with China: 0.21
  - with Euro Area: 1
  - with EMs: 0.05
  - with G10: 0.27
  - with Exported com.: 0.12
  - with Oil: 0.12
  - with VIX: -0.04
- Row: G10:
  - with Rand volatility: 0.24
  - with SAVID: 0.19
  - with SA: 0.52
  - with USA: 0.01
  - with China: 0.39
  - with Euro Area: 0.05
  - with EMs: 1
  - with G10: 0.07
  - with Exported com.: 0.04
  - with Oil: 0.34
  - with VIX: -0.10
- Row: Exported commodities:
  - with Rand volatility: 0.67
  - with SAVID: 0.25
  - with SA: 0.02
  - with USA: -0.18
  - with China: 0.09
  - with Euro Area: 0.27
  - with EMs: 0.07
  - with G10: 1
  - with Exported com.: 0.29
  - with Oil: 0.38
  - with VIX: 0.00
- Row: Oil:
  - with Rand volatility: 0.48
  - with SAVID: 0.06
  - with SA: 0.33
  - with USA: -0.04
  - with China: 0.11
  - with Euro Area: 0.12
  - with EMs: 0.04
  - with G10: 0.29
  - with Exported com.: 1
  - with Oil: 0.42
  - with VIX: -0.23
- Row: VIX:
  - with Rand volatility: 0.70
  - with SAVID: 0.40
  - with SA: 0.18
  - with USA: -0.14
  - with China: 0.42
  - with Euro Area: 0.12
  - with EMs: 0.34
  - with G10: 0.38
  - with Exported com.: 0.42
  - with Oil: 1
  - with VIX: 0.10
- Row: Political Uncertainty:
  - with Rand volatility: 0.14
  - with SAVID: 0.02
  - with SA: -0.11
  - with USA: 0.13
  - with China: -0.02
  - with Euro Area: -0.04
  - with EMs: -0.10
  - with G10: 0.00
  - with Exported com.: -0.23
  - with Oil: 0.10
  - with VIX: 1
- Notes from table:
  - Source: JSE, Citi, Bloomberg, Authors' calculations
  - Exported Commodities (includes Gold, Platinum, Coal and Iron ore) and Oil (Brent Crude) enter analysis expressed as daily price returns volatility
  - Citi ESIs (Macroeconomic Surprises), Global factors, Local Political Uncertainty

### Alternative measures of exchange rate volatility (Appendix II and Figure A1)
- Measures compared:
  - 3-month moving average standard deviation of the rand / U.S. dollar daily returns
  - SAVID
  - GARCH(1,1) — variance of rand / U.S. dollar daily returns
  - GARCH(2,1) — variance of rand / U.S. dollar daily returns
- Key methodological points:
  - Standard deviation (3-month moving average) provides a simple measure of historical volatility but assumes constant volatility over time.
  - Bollerslev’s GARCH models estimate time-varying standard deviation of daily returns; conditional variance depends on lagged disturbances and its own lagged values.
  - GARCH model equations (as presented):
    - yt = μ + ut
    - ut = εt ht 1/2, εt / It−1 ~ N(0,1)
    - ht = α0 + ∑_{i=1}^q αi u_{t−1}^2 + ∑_{j=1}^p βj h_{t−j}, t = 1, ....
  - Conditional variance nonnegativity requires α0 ≥ 0, αi ≥ 0 (i = 1,...,p), and βj ≥ 0 (j = 1,...,q).
  - Lag length selection based on information criteria; GARCH(2,1) suggested by information criteria, but GARCH(1,1) also reported for comparability.
  - Integrated GARCH (Engle and Bollerslev (1986)) and component GARCH (Engle and Lee (1999)) also estimated for robustness.
- Empirical comparison (Figure A1 summary):
  - All three volatility measures (SAVID, GARCH, 3-month moving standard deviation) follow broadly the same trend.
  - Common spikes at roughly the same times: early 2010 (euro zone debt crisis), late 2011 (rand trend appreciation reversed to depreciation), mid-2012 (euro zone contagion, falling commodity prices, Marikana massacre), early 2013 (taper tantrum), beginning of 2014 (second round of taper tantrum).
  - The 3-month moving standard deviation appears smoother compared to SAVID and GARCH but shows similar spikes.
  - Figure note: All series are set equal to 100 at the beginning.
  - Source for figure: JSE, Bloomberg, Author's own calculations

### Alternative GARCH models — estimation results (Table A3 and continuation)
- GARCH model outputs (selected reported coefficients and statistics)
- First reported model (table block):
  - C: 0.013, Std. Error: 0.005, z-Statistic: 2.694, Prob.: 0.007
  - RESID(-1)^2: 0.051, Std. Error: 0.008, z-Statistic: 6.155, Prob.: 0.000
  - GARCH(-1): 0.932, Std. Error: 0.0128, z-Statistic: 1.019, Prob.: 0.000
  - R-squared: 0.00
  - Adjusted R-squared: 0.00
  - Durbin-Watson stat: 2.00
  - F-statistic: 3.120, Prob.: 0.078
  - Obs*R-squared: 3.118, Prob.: 0.078
- Second reported model (GARCH(2,1) block):
  - C: 0.020, Std. Error: 0.007, z-Statistic: 2.978, Prob.: 0.003
  - RESID(-1)^2: -0.018, Std. Error: 0.020, z-Statistic: -0.899, Prob.: 0.369
  - RESID(-2)^2: 0.081, Std. Error: 0.024, z-Statistic: 3.374, Prob.: 0.001
  - GARCH(-1): 0.912, Std. Error: 0.0156, z-Statistic: 1.092, Prob.: 0.000
  - R-squared: 0.00
  - Adjusted R-squared: 0.00
  - Durbin-Watson stat: 2.00
  - F-statistic: 0.000, Prob.: 0.997
  - Obs*R-squared: 0.000, Prob.: 0.997
- Third reported model (IGARCH / alternative block):
  - RESID(-1)^2: 0.044, Std. Error: 0.005, z-Statistic: 8.364, Prob.: 0.000
  - GARCH(-1): 0.956, Std. Error: 0.0051, z-Statistic: 83.735, Prob.: 0.000
  - R-squared: 0.00
  - Adjusted R-squared: 0.00
  - Durbin-Watson stat: 2.00
  - F-statistic: 1.090, Prob.: 0.297
  - Obs*R-squared: 1.090, Prob.: 0.296
- Continued coefficient panels (Table A3/continued):
  - Panel 1 coefficients:
    - C(1): 0.773962, Std. Error: 0.112888, z-Statistic: 6.856016, Prob.: 0
    - C(2): 0.988369, Std. Error: 0.014986, z-Statistic: 65.95086, Prob.: 0
    - C(3): 0.034633, Std. Error: 0.061289, z-Statistic: 0.565069, Prob.: 0.572
    - C(4): 0.018755, Std. Error: 0.059549, z-Statistic: 0.314944, Prob.: 0.7528
    - C(5): 0.938349, Std. Error: 0.077605, z-Statistic: 12.09135, Prob.: 0
    - R-squared: 0.00
    - Adjusted R-squared: 0.00
    - Durbin-Watson stat: 2.00
    - F-statistic: 3.282638, Prob.: 0.0702
    - Obs*R-squared: 3.27995, Prob.: 0.0701
  - Panel 2 coefficients:
    - C(1): 0.752, Std. Error: 0.082, z-Statistic: 9.197, Prob.: 0.000
    - C(2): 0.974, Std. Error: 0.009, z-Statistic: 104.907, Prob.: 0.000
    - C(3): 0.061, Std. Error: 0.010, z-Statistic: 6.100, Prob.: 0.000
    - C(4): -0.121, Std. Error: 0.032, z-Statistic: -3.800, Prob.: 0.000
    - C(5): 0.072, Std. Error: 0.041, z-Statistic: 1.749, Prob.: 0.080
    - C(6): 0.077, Std. Error: 0.342, z-Statistic: 0.224, Prob.: 0.822
    - R-squared: 0.00
    - Adjusted R-squared: 0.00
    - Durbin-Watson stat: 2.00
    - F-statistic: 0.001, Prob.: 0.982
    - Obs*R-squared: 0.001, Prob.: 0.982
- Model forms and diagnostic notes (as presented):
  - IGARCH representation: GARCH = C(1)*RESID(-1)^2 + (1 - C(1))*GARCH(-1)
  - GARCH(2,1) representation: GARCH = C(1) + C(2)*RESID(-1)^2 + C(3)*RESID(-2)^2 + C(4)*GARCH(-1)
  - GARCH(1,1) representation: GARCH = C(1) + C(2)*RESID(-1)^2 + C(3)*GARCH(-1)
  - CGARCH / CGARCHT (component GARCH) representation:
    - Q = C(1) + C(2)*(Q(-1) - C(1)) + C(3)*(RESID(-1)^2 - GARCH(-1))
    - GARCH = Q + (C(4) + C(5)*(RESID(-1)<0))*(RESID(-1)^2 - Q(-1)) + C(6)*(GARCH(-1) - Q(-1))
    - Alternative CGARCH form: GARCH = Q + C(4) * (RESID(-1)^2 - Q(-1)) + C(5)*(GARCH(-1) - Q(-1))
  - Information criteria referenced: Akaike info criterion, Schwarz criterion, Hannan-Quinn criter.
  - Heteroscedasticity tests reported: ARCH LM test statistics (Prob. F(1,1563), Prob. Chi-Square(1))

### Key empirical takeaways
- Multiple volatility measures (SAVID, GARCH variants, 3-month moving standard deviation) show consistent spike timing associated with major global and local events (euro zone debt crisis, rand appreciation/depreciation reversals, falling commodity prices, Marikana massacre, taper tantrum episodes).
- GARCH specifications (including GARCH(1,1), GARCH(2,1), IGARCH, component GARCH) generally report statistically significant persistence parameters (high coefficients on GARCH(-1) in several specifications), indicating volatility persistence in rand / U.S. dollar returns.
- Stationarity tests (ADF and PP) show rejection of unit root null for most variables at conventional significance levels (p-values reported exactly per variable), supporting use of models that assume stationarity of the series reported.

*Source: _wp16205 - Appendix I—Descriptive Statistics and Detailed Results (JSE, Citi, Bloomberg, Authors' calculations)*

### Appendix III—Sensitivity Analysis - Robustness checks

### Appendix III—Sensitivity Analysis - Robustness checks

### A. Analysis using country grouping ESIs as explanatory variables
- Approach: Estimated specifications A through G using country-grouping ESIs (Advanced Economies (G10) and Emerging Markets (EMs)) versus the domestic ESI.
- Key empirical patterns:
  - Table A4, Column A: Rand volatility associated only with local macroeconomic surprises and surprises from Advanced Economies (G10) and selected EMs.
  - Columns B–D: Global volatility factors (South Africa’s main exported and imported commodity price volatility, and the VIX) have a large positive and significant effect on rand volatility; local political uncertainty also associated with increased rand volatility.
  - Column E (pooled specification): Rand volatility does not respond to local macroeconomic surprises, responds positively to G10 ESI and negatively to EMs ESI; increases in commodity price volatility, the VIX and local political uncertainty associated with increased rand volatility.
  - Column F (parsimonious): ESIs for South Africa excluded by F-test; results consistent with rand volatility mainly driven by international macroeconomic surprises, commodity price volatility and global market risk perception; local political uncertainty significant.
- Selected numeric results (Table A4):
  - Constants: 13.236*, 12.542*, 8.887*, 14.490*, 8.786*, 8.781*
  - SA ESI (#): 0.024*, 0.000
  - G10 ESI: 0.038*, 0.017**, 0.017**
  - EMs ESI: 0.018, -0.020**, -0.020**
  - Exported Commodities: 1.326*, 1.083*, 1.082*
  - Brent crude: 1.850*, 1.290*, 1.293*
  - VIX: 0.351*, 0.199*, 0.198*
  - Political uncertainty: 0.048**, 0.052*, 0.052*
  - R-squared (Columns A–F): 0.13, 0.54, 0.50, 0.02, 0.74, 0.74
  - Adjusted R-squared (Columns A–F): 0.13, 0.54, 0.50, 0.02, 0.73, 0.73
  - S.E. of regression (Columns A–F): 2.81, 2.04, 2.14, 2.99, 1.56, 1.55
  - Sample: 8/24/2009 8/24/2015, Included observations: 1566
  - HAC standard errors & covariance (Bartlett kernel, Newey-West fixed bandwidth = 8.0000)
  - Note: All Surprises (ESI) are in absolute values; * denotes significance at the 1%, ** at the 5% and *** at the 10% level of significance.

### B. Alternative measures of volatility
- Approach: Run specification F with alternative volatility measures (SAVID, Standard Deviation, GARCH (1,1), GARCH (2,2), Component GARCH with a trend, and Returns series).
- Main result: Findings robust across volatility measures—selected commodity price volatility, global market risk perceptions (VIX), and political uncertainty significantly drive rand volatility; macroeconomic surprises (USA, Advanced economies, EMs) influence rand volatility but with weaker/significance-variable evidence.
- Selected numeric results (Table A5: Specification E using alternative measures):
  - Constants: 8.511*, 0.369*, 0.037, 0.068, 0.074, -0.064*
  - SA ESI (#): 0.001, 0.001***, -0.001, -0.002**, -0.002**, 0.000
  - USA ESI: 0.013*, 0.002*, -0.001, -0.001, -0.001, 0.000
  - EU ESI: -0.004, -0.001***, -0.001, -0.001, -0.001, 0.001*
  - Exported Commodities: 1.034*, 0.092*, 0.136*, 0.131*, 0.125*, -0.020*
  - Brent crude: 0.992*, 0.101*, 0.121*, 0.119*, 0.114*, 0.036*
  - VIX: 0.221*, 0.010*, 0.026*, 0.025*, 0.026*, -0.003*
  - Political uncertainty: 0.048*, 0.005*, 0.006*, 0.006**, 0.006*, 0.001*
  - R-squared: 0.73, 0.74, 0.62, 0.56, 0.55, 0.31
  - S.E. of regression: 1.56, 0.12, 0.25, 0.27, 0.27, 0.07
  - Sample: 8/24/2009 8/24/2015, Included observations: 1566
- Selected numeric results (Table A6: Country groupings with alternative volatility measures):
  - Constants: 8.786*, 0.451*, 0.168*, 0.184**, 0.186*, -0.089*
  - G10 ESI: 0.017**, 0.001***, 0.000, 0.000, 0.000, 0.001
  - EMs ESI: -0.020**, -0.003*, -0.007*, -0.007*, -0.007*, 0.001*
  - Exported Commodities: 1.083*, 0.106*, 0.152*, 0.144*, 0.137*, -0.022*
  - Brent crude: 1.290*, 0.124*, 0.092*, 0.086*, 0.081*, 0.036*
  - VIX: 0.199*, 0.007*, 0.028*, 0.028*, 0.028*, -0.003*
  - Political uncertainty: 0.052*, 0.007*, 0.006*, 0.005*, 0.005*, 0.001***
  - R-squared (columns): 0.74, 0.75, 0.64, 0.57, 0.56, 0.17
  - S.E. of regression (columns): 1.56, 0.12, 0.25, 0.27, 0.27, 0.08
  - Sample: 8/24/2009 8/24/2015, Included observations: 1566

### C. Robustness depending on whether the rand is appreciating or depreciating
- Approach: Specification F augmented with a dummy interaction equal to 1 when the rand is appreciating.
- Main findings:
  - Regardless of appreciation/depreciation, increases in surprises from the United States, exported and imported commodity price volatility, the VIX and local political uncertainty increase rand volatility.
  - When the rand is depreciating, an increase in the VIX leads to greater volatility compared to when the rand is appreciating.
- Selected numeric results (Table A7):
  - Constant: 8.538*
  - USA ESI: 0.014*
  - Exported Commodities: 1.029*
  - Brent crude: 1.086*
  - VIX: 0.196*
  - Political uncertainty: 0.056*
  - Interaction terms: USA ESI -0.001; Exported Commodities 0.026; Brent crude -0.188; VIX 0.033**; Political uncertainty 0.001
  - R-squared: 0.73; Adjusted R-squared: 0.73; S.E. of regression: 1.56; Sum squared resid: 3794.82; Log likelihood: -2915.10
  - Sample: 8/24/2009 8/24/2015, Included observations: 1566

### D. Asymmetry between positive and negative surprises (ESIs)
- Question: Whether positive (good) and negative (bad) surprises from the United States have asymmetric impacts on rand volatility.
- Method: Added interaction between |ESI_US,t| and dummy D_ESI_US,i equal to 1 for positive US ESI values.
- Result: Rejected asymmetry—the dummy interaction coefficient statistically insignificant; both good and bad United States ESIs increase rand volatility equally.

### E. Lagged dependent
- Approach: Added lagged dependent variable (SAVID (-1)) to preferred specification as sensitivity check.
- Main findings:
  - Volatility is persistent: high (low) volatility followed by high (low) volatility.
  - Coefficient on lagged volatility positive and statistically significant; many surprise indices and commodity volatility coefficients become insignificant, but global volatility effect remains statistically significant.
- Selected numeric results (Table A8):
  - Constants (A–E): 0.241*, 0.334**, 0.299*, 0.238*, 0.497*
  - Exported Commodities: 0.027***, 0.0452*
  - Brent crude: 0.068**, 0.034
  - VIX: 0.026*, 0.030*
  - SAVID (-1): 0.981*, 0.971*, 0.950*, 0.985*, 0.931*
  - R-squared (A–E): 0.97 (all)
  - S.E. of regression (A–E): 0.51, 0.51, 0.50, 0.51, 0.50
  - Sample: 8/24/2009 8/24/2015, Included observations: 1566

### F. Time-varying coefficients
- Approach: Re-estimate preferred specification F using a recursive least squares framework to detect changes over the sample.
- Main findings:
  - Responsiveness of rand volatility to macroeconomic surprises, commodity price volatility, and global volatility has changed over time.
  - No single common structural break identified.
  - Notable development: Surprises from the United States shifted from reducing rand volatility in the initial sample phase (2009/2010) to increasing volatility since mid-2013—coincident with tapering of quantitative easing.
  - Interpretation: Post-global financial crisis negative US surprises initially increased uncertainty and pushed investors to emerging markets; during quantitative easing/tapering periods, positive US surprises are “bad” for South Africa as funds relocate to safer assets.

*Source: Authors' calculations, Appendix III—Sensitivity Analysis - Robustness checks.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16205.pdf_