Insights into Gauteng's Leading Composite Indicator Study

Gauteng Leading Composite Indicator by MW Hempson, DT Makhubela & GV N lting Gauteng Provincial Treasury November 2013 Presented by Geoff N lting 1

Contents Introduction Literature and Background South African Overview Gauteng Province s Economic Performance Data and Methodology Statistical Problems and Future Research Conclusion 2

Introduction LCIs are useful tools for policymakers and predict the direction of changes in economic activity, measured by the business cycle, over the short-term. LCIs and other composite indicators have been compiled for developed nations and have been used extensively for policy making. Not yet available for most emerging markets and developing countries. This study made use of econometric techniques to create a LCI for the Gauteng provincial economy. Variables being tested for inclusion in the model were tested against the Gauteng GDP-R. 3

Introduction There has been little research about LCIs in developing countries (EMEs) because: 1. EME data is usually limited in availability and high frequency data is sporadically accessible. 2. The business cycles in these countries are more dependent on weather conditions, due to the reliance on the primary sector. 3. EMEs are usually prone to sudden crises, which makes it difficult to distinguish business cycles. 4

Literature and Background These indicators were first pioneered by Burns and Mitchell (1946) in the context of the US economy. The OECD publishes leading indicators on a monthly basis for its member countries since 1987. Moore and Shiskin (1967) developed a formal weighting system by scoring the variables using their economic significance, statistical adequacy and cyclical turning points. Short-coming; weightings associated with them were based on a subjective economic analysis and not a scientific econometric model. 5

South African Overview In SA, the construction of the leading indicator, is done on a monthly basis by SARB. The SARB first constructed its LCI in 1983. Factors such as the structural changes in the economy, new economic indicators or discontinuation of existing variables, led to frequent reassessment of the indicator. The components of the LCI were also evaluated according to their economic significance, statistical adequacy, historical reaction with the business cycle and the timeless nature of the data. The SARB does not only compile the leading indicator but, also the coincident and the lagging indicators. Both these leading indicators are used to predict the turning points of the reference indicator usually measured by GDP. 6

South African Overview SARB LCI Components Variables Indicators Average hours worked per factory worker in manufacturing (half weight) BER Business Confidence Index Volume of orders in manufacturing (half weight) Of South Africa s main export commodities (US Commodity Price Index dollar based) South Africa s major trading-partner countries Composite Leading Business Cycle Indicator (percentage change over 12 months) Gross Operating Surplus As a percentage of gross domestic product Index of Prices Prices of all classes of shares traded on the JSE 10-year government bonds minus 91-day Treasury Interest Rate Spread bills As appearing in the Sunday Times newspaper Job Advertisements (percentage change over 12 months) Number of Building Plans Approved Flats, townhouses & houses larger than 80m Number of New Passenger Vehicles Sold Percentage change over 12 months Real M1 Six-month smoothed growth rate Source: SARB, 2013 Note: Half weights are allocated to opinion-based surveys. 7

South African Overview GDP and LCI, South Africa, 2001Q1-2013Q1 3 10 2.5 8 2 6 1.5 4 1 2 0.5 0 0 -2 -0.5 -4 -1 -6 -1.5 -8 2007/01 2001/01 2001/03 2002/01 2002/03 2003/01 2003/03 2004/01 2004/03 2005/01 2005/03 2006/01 2006/03 2007/03 2008/01 2008/03 2009/01 2009/03 2010/01 2010/03 2011/01 2011/03 2012/01 2012/03 2013/01 The GDP of the country tracked the trend of the LCI over the review period with the LCI leading the turning points. The LCI reached recession in 2008 before the GDP following in 2009 and the recovery thereafter. The LCI also lead the decline in the last quarter of 2010 as world economic uncertainty looms, especially in the Euro area. 8 Source: Quantec Research, 2013

Gauteng Provinces Economic Performance Gauteng accounts for the highest contribution to the national economy of approximately 35.9 percent of the country s GDP, followed by KwaZulu-Natal at 16.4 percent and Western Cape at 14.8 percent. According to Economist.co.za, the Provincial Economic Barometers are instruments used to measure economic activity levels for the various provinces and consist of a set of sub-indexes that measure the performance of the province as well as individual economic sub- sectors. Amongst others, the indexes include the growth, trade, and economic stress index. The growth index is compiled through the consolidation of the performance of all the individual economic sector indexes. The trade index uses information from the wholesale and retail trade as well as tourism and entertainment. 9

Data and Methodology Definition of Variables and Sources of Data Name Definition Recorded building plans passed for residential dwelling houses 802m Sources Quantec Stats SA BLD in GP Civil cases recorded in large magistrates offices in GP Quantec Stats SA CIV Quantec IMF IFS CPI Consumer price index for SA Quantec IMF IFS ER Rand/US$ exchange rate for SA FN15 Financials 15 Index for SA McGregor BFA GDP-R Gross Domestic Product by region for GP Quantec Quantec Absa HSE Middle class houses purchase price in GP Quantec IMF IFS M1 M1 money supply in SA Quantec SAPIA PET93 Price of 93 octane fuel (lead replacement) in GP Quantec NAAMSA VHCL Total Vehicle Sales in GP In order to develop the Gauteng LCI, selected indicators considered for the study were visually compared to GDP-R before testing for significance. Notes: Stats SA = Statistics South Africa, IMF IFS = International Monetary Fund s International Financial Statistics, SAIPA = South African Petroleum Industry Association, NAAMSA = National Association of Automobile Manufacturers of South Africa. 10

Data and Methodology The sample of data comprises of quarterly observations of 10 variables from various sectors of Gauteng and South African economies, from the first quarter of 1998 to the fourth quarter of 2012 (60 observations). Monthly data broken down to the provincial level is not available for many of the indicators under study. The US$ exchange rate and the M1 money supply indicators were included because of the large financial sub-sector in the province. Due to the lack of availability of provincial data, nominal variables were not deflated by the South African CPI as the national CPI was used as an indicator. 11

Data and Methodology According to economic theory, a decline in the number of building plans passed is, ceteris paribus, followed by a corresponding decrease in the GDP-R of the region. An increase in civil cases of debt may be taken as an indication that the economy is slowing down and people are unable to honour their debt obligations, as they may have lost their jobs or their business is no longer profitable. The CPI was also selected to capture the effect of price increases to the province s economy. The provincial CPI data only goes back as far as 2008, thus the South African CPI was tested as a proxy. House prices were also selected as another variable that can have an effect on the province s economy because the increase in house prices signify increased consumer confidence. 12

Data and Methodology The exchange rate was included as one of the indicators that could be used to construct the Gauteng LCI, as the province accounts for the largest proportion of the country s international trade. According to Quantec data, the province accounted for approximately 67.3 percent and 60.1 percent of the country s export and imports respectively, in 2012. Exports to China (13.7 percent), USA (9 percent) and Japan (6.7 percent). Imports from China (15.6 percent), Germany (12.1 percent) and USA (9.9 percent). Other indicators that were considered for this study include the Gauteng petrol price and vehicle sales in the province. 13

Data and Methodology Financial Sub-Sector's Contribution to GVA-R Total, SA & GP, 2003-2016* 27 26 25 24 23 % 22 21 20 19 SA GP The financial & business services contributes the second largest proportion to GVA and GVA-R as well. The Financial 15 index of the Johannesburg Stock Exchange was considered for inclusion because of this reason. 14 Source: IHS Global Insight, 2013

Data and Methodology Seasonal Adjustment Seasonally Adjusted Petrol Price, Gauteng, 1998-2012 14 12 10 8 R 6 4 2 0 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 PET93 PET93_SA 15 Source: Quantec Research & GPT own calculation, 2013

Data and Methodology Unit Root Tests Variables should be time-stationary in order to be included in ordinary least squares models (OLS). This is done to prevent spurious correlations associated with non-stationary variables. All seasonally adjusted variables were then tested in level and logarithmic form for non-stationarity using the ADF. The variables were then transformed into annual growth rates and tested again for unit roots. ADF tests are known to have size and power problems when conducted in small samples. These variables were re-tested using the Phillips-Perron test statistic, as it performs better in smaller samples than the ADF test. 16

Data and Methodology Unit Root Tests Test ADF Phillips-Perron Transformation Level Logarithm Annualised Annualised Series t-Value t-Value t-Value t-Value BLD -1.44 -1.39 -1.80 -2.83 * CIV -1.16 -1.06 -0.77 -4.56 *** CPI 1.27 -0.11 -2.38 -2.70 ** ER -2.91 * -2.88 * -2.15 -2.80 * FN15 0.46 0.01 -1.83 -3.02 ** GDP-R 0.75 -0.53 -1.74 -2.84 ** HSE -1.07 -2.02 -1.23 -1.39 M1 1.87 -1.47 -1.97 -2.93 * PET93 0.86 -0.46 -2.51 -3.45 ** VHCL -0.86 -0.93 -1.88 -2.54 Source: GPT own calculation, 2013 Notes: Annualised data of logarithm data. *, **, *** denote significance at 10, 5 and 1 percent level, respectively. indicates significance at the 12 percent level. Critical Values according to MacKinnon (1996). 17

Data and Methodology Correlation Analysis Ordinary Correlations GDP- Indicator BLD CIV CPI ER FN15 M1 PET93 VHCL R GDP-R 1.00 BLD 0.49 1.00 CIV -0.12 -0.25 1.00 CPI -0.14 -0.48 0.06 1.00 ER -0.02 -0.21 0.04 0.34 1.00 FN15 0.48 0.48 -0.14 -0.65 -0.32 1.00 M1 0.59 0.18 -0.01 -0.17 0.21 0.33 1.00 PET93 0.64 0.30 -0.35 0.11 0.17 0.19 0.31 1.00 VHCL 0.53 0.78 -0.18 -0.75 -0.33 0.72 0.20 0.35 1.00 18 Source: GPT own calculation, 2013 Notes: Ordinary correlations of logged annualised indicators.

Data and Methodology Granger-Causality Tests It is of utmost importance when conducting research on LCIs to determine whether one series actually leads another specifically, that the indicator leads the reference series. This is imperative so that reliable indicators are identified. The pairwise Granger-Causality test was developed for such a purpose. This test works by analysing whether changes in the indicator series are followed by changes in the reference series, and vice versa. 19

Data and Methodology Pairwise Granger-Causality Tests H0: Indicator not Granger-causal H0: GDP-R not Granger-causal Result Indicator F-Statistic F-Statistic BLD 2.27 * 4.36 *** Feedback CIV 1.21 0.54 No Causality I GDP-R CPI 3.42 ** 0.37 ER 0.59 1.42 No Causality FN15 3.49 ** 2.15 * Feedback M1 3.63 ** 2.57 ** Feedback I GDP-R PET93 2.28 ** 1.91 GDP-R I VHCL 1.21 2.88 ** Source: GPT own calculation, 2013 Notes: *, **, *** denote significance at 10, 5 and 1 percent level, respectively. VAR lag-length equal to 4 for all indicators. 20

Data and Methodology Model Estimation This study followed the methodology of the OECD (1987), whereby the leading horizon is set to two quarters (six months). This research employs a simple variable selection criterion and a linear reduced form regression equation. Initially, the generalised model made use of all available indicators but indicators were systematically removed based on the variable that had the lowest t-ratio. Multicolinearity was also avoided. In order to determine the statistical relationship between the LCI and GDP-R, GDP-R is shifted two quarters ahead. 21

Data and Methodology Model Estimation The final model is represented by the reduced form in the equation below. 4lnGDP_Rt+2= + 4lnLCIt + t t = t+ t-1 Where; 4lnGDP_Ris the annualised growth rate of the seasonally adjusted logarithmic transformation of GDP-R for the Gauteng Province; 4lnLCI is a vector of seasonally transformations of coincident indicators expressed in annual growth rates; and t is an error term. MA(1)is a first order moving average component of the error term. adjusted logarithmic The errors were tested for heteroskedasticity using the autoregressive conditional heteroskedasticity (ARCH). The test indicates that the errors are homoskedastic. 22

Data and Methodology Model Estimation The above variable selection procedure identified a constant, five coincident indicators and a significant first order moving average component. Rand/US$ exchange rate (ER), the Financials 15 index (FN15), M1 money supply (M1), petrol price (PET93) and total vehicle sales in Gauteng (VHCL). All variables are significant at approximately the 15 percent level of significance and lower and explain 88 percent of the variation in the growth of the GDP-R of the Gauteng economy. 23

Data and Methodology Model Estimation Estimation of Leading Composite Index Dependent Variable: GDP_R(+2) Method: Least Squares Variable C ER FN15 M1 PET93 VHCL MA(1) Coefficient 2.38 -0.02 0.02 0.14 -0.02 0.04 0.96 Std. Error 0.31 0.01 0.01 0.02 0.01 0.01 0.04 t-Statistic 7.66 -1.81 1.59 5.88 -1.47 3.49 24.52 Prob. 0.00 0.08 0.12 0.00 0.15 0.00 0.00 All variables significant at the 10 percent level, except the Financials 15 Index and petrol price, which are only significant at the 12 percent and 15 percent levels, respectively. Due to the limited number of variables, the FN15 and PET93 variables were retained in the model. 24 Source: GPT own calculation, 2013

Data and Methodology Model Estimation Estimation of LCI, Actual, Fitted and Residual Values, 1999-2012 4 8 6 3 Actual & Fitted 4 2 Residual 2 1 0 -2 0 -4 -1 -6 -2 -8 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Residual Actual Fitted The fitted values are used to create a LCI by setting the initial observation of the LCI equal to the equivalent observation of GDP-R. The LCI derived is illustrated in the following figure, and closely resembles the behaviour of the GDP-R. The most striking result is that the LCI offers a comparatively accurate forecast of most turning points in the GDP-R two quarters before the actual GDP-R observation. 25 Source: GPT own calculation, 2013

Data and Methodology Model Estimation Comparison of LCI and GDP-R, 1999-2012 8 6 4 2 0 -2 -4 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 GDP-R LCI The performance of the LCI, based on the above results, successfully indicates the direction of the economy and can thus provide advice to policymakers. However, it is recommended that these results should be utilised with caution as this LCI should only be used simultaneously with appropriate qualitative information, for short-term policy decisions. 26 Source: GPT own calculation, 2013

Statistical Problems and Future Research The main statistical problems encountered in carrying out this research were: 1. Mostly low frequency data was available. 2. Data at a provincial level was scarcely obtainable. 27

Statistical Problems and Future Research Forecast the LCI by means of recursive methods. Make use of higher frequency data (if available), or by converting quarterly data using the cubic spline interpolation process into monthly data. And then adjusting the monthly data final cyclical components by means of the Months for Cyclical Dominance (MCD) method. Experiment with different filters to seasonally adjust the data. Use different weighting methods. Conduct in-house qualitative surveys in Gauteng as Thompson and Walstad (2012) have done when developing a LCI for the state of Nebraska. Use other indices from other countries, as Mongardini and Saadi-Sedik (2003) did when estimating the LCI for Jordan, for indicators of the global business cycle. Note: Red text indicates what is currently being done in the GPT revised model. 28

Statistical Problems and Future Research Create a reference coincident series that the LCI can forecast. Test for causality using individual Granger-causality tests as done by Fritsche and Marklein (2001). Make use of the Bry-Boschan methodology to identify turning points. As time passes, update and judge the performance of this study s model as a larger sample will provide more accurate results. Make use of ARIMAX and error correction models to make GDP- R forecasts (Kl ucik and Juriov , 2010). Go further than identifying the direction of the business cycle by making use of a dynamic factor model to predict the magnitude of the growth rate of GDP-R (Chauvet et al, 2000). 29 Note: Red text indicates what is currently being done in the GPT revised model.

Conclusion This study outlined the initial challenges of the earlier models, which were based on subjective rather than econometric processes for variables that were comprised in the composite indexes of the leading indicator. Challenges of data availability in most of EMEs remain eminent, especially at regional levels. However, it is still possible with limited data observations, to establish a meaningful correlation between economic indicators and forecast the future direction of economic variables. This study developed a formal statistical approach to investigate whether the GDP-R of the province can be forecast by an LCI. A simple linear regression model was proposed for this purpose. 30

Conclusion The results indicate that the LCI seems to provide an index that reflects the reference series, GDP-R, business cycle fairly well, even with a limited number of series (indicators) and a relatively short time horizon available. However, verification of the LCI s reliability and predicative ability requires a long-term experimental application, including many revisions as confirmed by the experience of many other countries. The authors caution that the results of this study should only be used to gauge the direction of the economy over the short- term. While the LCI can pinpoint the state (direction) of the economy in the cycle, it is incapable of forecasting the precise magnitude of the economic activity. This is left for future research. 31

End Thank You 32

Additional Information For further information contact Gauteng Provincial Treasury, 75 Fox Street, Imbumba House, Marshalltown, 2107. Tel: 011 227 9000 Fax: 011 227 9055 Email: GPTCommunication@gauteng.gov.za Economic Bulletin document available online from: http://www.treasury.gpg.gov.za/Document/Documents/Economic %20Bulletin%20Quarter%201%20-%202013-14.pdf 33