Presentation on theme: "Economic Crises and Forecasting: a review of South Africas model by P. Lehohla and D. Morudu Policy Research and Analysis Unit Office of the Statistician."— Presentation transcript:
Economic Crises and Forecasting: a review of South Africas model by P. Lehohla and D. Morudu Policy Research and Analysis Unit Office of the Statistician General Statistics South Africa Paper prepared for the 15 Conference of Commonwealth Statisticians held on 7 – 10 February 2011 First draft: 26 January 2011 This draft: 3 February 2011 Draft remains preliminary and incomplete Enquiries: Hlabim@statssa.gov.zaHlabim@statssa.gov.za
Table of contents Introduction Approach in South Africa A general problem: evidence from South Africa Literature review Some proposals: the potential role for Stats SA Conclusions
Introduction Economists have useful scientific tools that enable them to anticipate regular business cycles. Indeed economists provide valuable input in shaping monetary/fiscal policies based on anticipated business cycle movements. What economists often fail to do is to anticipate major economic crises despite an accumulation of data series in the financial sector, balance of payments and national budgets over the years, including the international tightening of data standards.
Introduction Following the 2008 global crisis, a number of forums consisting of expert academia and practitioners, sponsored by international agencies such as the International Monetary Fund, World Bank and United Nations, have been developed to review the current state of economic forecasting. An impressive number of papers on the state of current economic forecasting have surfaced.
Introduction Recent research reveals efforts to improve three major approaches, viz. (a)the composite business cycle indicators approach as developed at the National Bureau of Economic Research (NBER) in the United States (b)econometric modeling approach following recommendations of the Cowles Commission (c)hybrids of both the business cycle indicator approach and econometric approach, more notably illustrated through Markov switching autoregressive techniques
Introduction A further development consists of empirical studies that are based on the behavior of business cycles in a number of countries and how the business cycles interact, i.e. along the lines researched by the Center for International Business Cycle Research at Rutgers University.
Introduction South Africa currently produces, through the South African Reserve Bank, composite business cycle indicators using the NBER approach, and publishes a composite leading indicator index on a monthly basis. Other approaches have been suggested, but remain outside existing official economic and statistics agencies, and have not been pursued (e.g. E. Moolman (2004) on a Markov switching model for South Africa).
Introduction The objective of this paper is to assess weaknesses in South Africas business cycle forecasting efforts given current international developments in business cycle research.
Approach in South Africa To anticipate business cycle movements, and potentially economic crises, South Africa produces a set of composite business cycle indicators along the lines proposed by the NBER. A selection of time series is made from a vast number of economic time series and categorized into three distinct groups, viz. (a) leading indicator series; (b) coincident indicator series, and (c) lagging indicator series. The movement of each composite indicator is a weighted average of movements of a number of variables (Venter and Pretorius, 2004; Venter, 2004).
Table 1 Components of composite business cycle indicators Composite coincident indicatorComposite leading indicator 1. Gross value added at constant prices, excluding agriculture, forestry and fishing 1. Net balance on manufacturers observing an increase in the volume of orders received (half weight) 2. Value of wholesale, retail and new vehicle sales at constant prices 2. Number of new passenger vehicles sold (year on year percentage change) 3. Utilization of production capacity in manufacturing 3. Opinion survey of business confidence: manufacturing, construction and trade 4. Total formal non-agricultural employment 4. Composite leading business cycle indicator of major trading partner countries (year on year percent) 5. Industrial production index 5. Index of commodity prices (in US$) for a basket of South African export commodities Composite lagging indicator 6. Real M1 money supply (deflated with the CPI, 6 months smoothed growth rate) 1. Value of non-residential buildings completed (constant prices)7. Index of prices of all classes of shares traded on the JSE 2. Ratio of gross fixed capital formation in machinery and equipment to final consumption expenditure on goods by households 8. Number of residential building plans passed for flats, townhouses and houses larger than 80m 2 3. Ratio of inventories to sales in the manufacturing and trade sectors9. Interest rate spread: 10 year bonds less 91 day Treasury bills 4. Nominal labor costs per unit of production in the manufacturing sector (% change over 4 quarters) 10. Gross operating surplus as a percentage of GDP 5. Cement sales in tons 11. Job advertisement space in the Sunday Times newspaper (% change over twelve months) 6. Ratio of households use of installment sale credit to their disposable income 12. Net balance of manufacturers observing an increase in the average number of hours worked per factory worker (half weight) 7. Predominant prime overdraft rate of banks
Approach in South Africa Continued increases in e.g. the composite leading indicator, for periods of up to six months suggest the gross domestic product (GDP) is likely to increase in subsequent months (Venter, 2005). Similarly, continued decreases in the composite leading indicator, for about two quarters, suggest the GDP is likely to decrease in subsequent quarters.
A general problem: evidence from South Africa In retrospect, the composite leading indicator could have provided, as early as August 2006, an sign that the economy was headed towards a recession The composite leading business cycle indicator decreased almost consistently from 127,5 in July 2006 to reach 114,9 in August 2008, i.e. about 25 months in advance
A general problem: evidence from South Africa The release of the composite leading business cycle indicator in South Africa is hardly topical: –it is released ex post, often after critical ex ante economic decisions have been made –the composite leading indicator is released monthly, almost 6 to 8 weeks after the reference month –the initial set of published data is revised for up to three months before being considered final
A general problem: evidence from South Africa When the composite leading business cycle indicator began declining in August 2006, initial estimates became available around October 2006 and only become final around January 2007, i.e. a 5-month lag. By December 2007, i.e. at the official peak of the business cycle, authorities were considering the final results of a very moderate decrease of the composite leading indicator from 125,4 in June 2007 to 125,3 in July 2007.
A general problem: evidence from South Africa There are even more significant delays in the determination of business cycle turning points –South African business cycles are determined as a deviation from a long-term trend, turning points are confirmed ex post after a significant number of months (Venter, 2005). –In terms of the SARB Quarterly Bulletin, the official business cycle peak, i.e. December 2007, was published in September 2009, almost two years after the fact.
A general problem: evidence from South Africa The significance of the composite leading business cycle indicator in policy formulation is not readily apparent –while the composite leading indicator suggested a pending downward economic phase through its movement from August 2006 to July 2007, monetary authorities continued to raise the repo rate in 9 subsequent Monetary Policy Committee meetings from 7,5% in 2006:06 to 12% in 2008:06.
In brief –the composite leading business cycle indicator lacks a mechanism to signal pending economic crises, it does not distinguish between regular recessions and crises –The composite leading business cycle is released with lags that are rather too long for policy makers to maximize the indicators potential usefulness. –These problems are not uniquely South African, but are typical challenges to the component business cycle indicator approach.
Literature review Predicting economic crises remains a major challenge in economics –Research suggests the existence of significant information rigidities which make it difficult for forecasters to promptly incorporate useful domestic and international news [D. Harding, et al. (2010), K. Carstensen, et al. (2010), P. Loungani, et al. (2010), D. Bragoli (2010), K. Drechsel, et al. (2010)]
Literature review Recent literature reveals efforts to improve three major approaches, viz. (a)the composite business cycle indicators approach (b)econometric modeling approach (c)hybrids of both the business cycle indicator approach and econometric approach
Literature review On business cycle indicators, a number of studies propose refined techniques for selecting leading, coincident and lagging composite indicators The selection of more refined indicators, including qualitative indicators, is meant to provide more responsive composite indicators with more useful business cycle or economic crisis signaling qualities. [G. Cubadda, et al. (2010), F. Youssef, et al. (2010), G. de Bondt, et al. (2010)]
Literature review A number of studies propose methods to quicken the production and release of business cycle indicators, including tentative signals that distinguish between prospective recessions and crises [Z. Guohua (2010), H. Lee, et al. (2010)] –E.g. the Business Cycle Signal System in China, or the Business Cycle Clock in Korea, have in-built critical values that help inform markets immediately whether the economy is overheated, likely to overheat, normal, likely to cool down or cool.
Literature review Further proposals to facilitate prompt supply of data for business cycle composite indicators include nowcasting [M. Pedersen (2010), E. Andreou, et al. (2009), K. Lee, et al. (2010), M. Wildi (2009), M. Camacho, et al. (2010)] –Pedersen (2010) suggests the development of monthly indicators of economic activity to approximate monthly GDP as practices in a number of Latin American countries –Andreau, et al. (2009) suggest the use of daily financial data to estimate current real economic activity or live GDP
Literature review A second set of proposals consists of calls for the development of more rigorous econometric models in business cycle forecasting [M. Franchi, et al. (2010), P. Exterkate, et al. (2010), S. Grassi, et al. (2010), T. Clark (2010), P. Foschi, et al. (2010), M. Lupinski (2010), O. Biau, et al. (2010), L Bisio, et al. (2010), M. Chauvet, et al. (2010), L. Lemoine, et al. (2010)]
Literature review Chauvet, et al. (2010), construct a set of leading business cycle indicators using a time-varying autoregressive probit model. Grassi, et al. (2010) and Clark (2010) propose the use of Bayesian Vector Autoregressive models using individual economic variables rather than composite business cycle indicators. Lupinski (2010) suggest use of a dynamic factor model using data with mixed frequencies and ragged edges. Lemoine, et al. (2010) suggest use of stochastic volatility in the mean (SV-M) models to anticipate business cycles.
Literature review Other econometric proposals include: –Exterkate, et al. (2010) who propose the use of kernel ridge regressions, used extensively in machine leaning communities, to estimate non-linear and high-dimensional business cycle series relations using kernels. –Biau, et al. (2010) propose a new business cycle technique that is based on the random forest, currently in use in medical and biological research.
Literature review A third set of proposals consists of an admixture between the composite indicator approach and the econometric approach. –The proposals are concerned with the linear features of the business cycle composite indicator weights, and suggest introducing non linear properties in the prediction of business cycles / crises [E. Learner, et al. (2002), S. Altug, et al. (2010), S. Senyuz, et al. (2010)]
Literature review Arguably, dynamics differ between recessions and expansions, e.g. recessions have larger shocks than expansions, and expansions have much wider durations than recessions. To accommodate the non-linearity of business cycles, studies suggest use of Markov switching autoregressive models. An added feature of the proposal is that the approach is more transparent than the NBER approach (i.e. there is no need for private Business Cycle Dating Committee meetings), and results are easily reproducible.
Literature review Further research consists of empirical studies that seek to capture the behavior of business cycles in a number of countries and how the business cycles interact, i.e. along the lines researched by the Center for International Business Cycle Research at Rutgers University [S. Altug, et al. (2010), R. Male (2010), M. Antonakakis, et al. (2010), J. Allegret, et al. (2010), U. Bergman, et al. (2010), J. Fidrmuc, et al. (2009), J. Goggin, et al. (2010)]
Literature review The latter set of studies provides a promising basis for developing a framework for anticipating local, regional or global economic crises. The studies also indirectly highlight challenges with data from diverse countries, diverse stages of statistical development among countries on business cycle analysis and structures.
Literature review The literature is also abundantly explicit on the challenges of market innovations that undermine proper monitoring of the economy –the development of new financial instruments [mortgage backed securities (MBS), collateralized debt obligations (CDO), credit default swaps (CDS)] in shadow banking and off balance sheet mechanisms, eluded normal economics and banking surveys and regulations [C. Towe (2009), H. Remsperger (2008), W. Bier (2009)] –That is, it remains imperative that statistical agencies monitor developments in the economy and continually identify possible improvements in the collection of data on new and modified market instruments.
Some proposals: potential role of Stats SA Statistic South Africa provides most of the series used for the development of the composite business cycle indicators –One possible area for the increased involvement of Stats SA is through undertaking required opinion surveys. Most of the opinion surveys in South Africa are conducted outside Stats SA, e.g. private sector.
Some proposals: potential role of Stats SA Phenomena like shadow banking and off balance sheet mechanisms, if these were prevalent in South Africa, would have completely eluded inclusion in business cycle indicators and hence proper monitoring. –It remains imperative that Stats SA constantly monitor innovative developments in the economy and constantly identify possible improvements in the collection of data on modified and new market instruments.
Some proposals: potential role of Stats SA The South African business cycle approach seems to fare well in anticipating business cycles, but does not provide adequate signals to anticipate economic crises. The approach is also unable to provide direction on time, with delays of about 5 months to two years
Some proposals: potential role of Stats SA This study suggests (along proposals already made by statistics agencies of China, Korea, the Netherlands and others) the development of economic activity indicators that can be produced instantly with the release of any relevant new data, and provide an instant barometer of the current and prospective state of the economy The release of more monthly indicator time series, rather than quarterly series, would complement the approach Also, proposals such as nowcasting, meant to provide economic foresight have to be natured and fully explored
Some proposals: potential role of Stats SA Another area that needs constant attention, at least among statistical agencies, is the continued development of statistics across countries. –Not all countries have clear business cycle indicator programs, many indicators are not seasonally adjusted, not deflated, not continuous, and do not have comparable frequencies. –There is need for an organized effort to develop business cycle indicators internationally.
Conclusions The study outlined South Africas main business cycle forecasting approach, viz. the composite business cycle approach, and identifies the major shortcomings of the approach as: (a)indicators are developed retrospectively with doubtful prospective inputs (b)indicators do not provide useful signals to distinguish between prospective economic downturns and crises (c)market innovations as depicted in shadow banking and off balance sheet mechanisms would not have been captured through current composite business cycle indicators
Conclusions The paper explores recent literature on business cycle forecasting and presents tentative proposals to (a)quicken the production of composite indicators (b)devise mechanisms that prospectively signal between a simple economic downturn and a crisis (c)guard against market innovations that make monitoring difficult (d)improved upkeep and processing of data internationally.
Conclusions The area of what approach to adopt, especially among the three major approaches, remains unclear. –The NBER approach is quite entrenched, historically and in terms of usage across countries. The approach however posits significant challenges in terms of transparency and reproducibility of results –Other proposed approaches that are more transparent with easily reproducible results, may develop to overtake the dominant NBER approach, especially if the current NBER data/infrastructure remains useful.