New Coincident and Leading Indexes for the Lebanese Economy

1 Introduction

Lebanon’s weak economic statistics are impeding timely decision making by businesses, investors and policy makers. Over the years, the quality of economic statistics in Lebanon has been extremely weak in terms of data compilation and frequency, even relative to countries with similar level of development (IMF 2012). Statistical weaknesses include areas such as national accounts, balance of payments, prices and inflation, and labor and social measures. In particular, as of May 2014, Lebanon’s latest GDP data are from 2011 and are only available on an annual basis. ^[1]

To provide a more timely and comprehensive assessment of economic activity in Lebanon, Banque du Liban (BdL) and the International Institute of Finance (IIF) separately developed coincident indexes. The former, which we denote by “BdL-CI”, was developed in 1993, immediately following the end of the civil war, and is composed of eight variables. ^[2] Notwithstanding the profound structural changes in Lebanon’s economy that took place since the end of the civil war, the weights of the eight variables in the BdL-CI have remained fixed since 1993. ^[3] Another caveat of the BdL-CI is the absence of a variable that captures the dynamics of the public administration sector, which represented on average 9.6 % of GDP between 2004 and 2011. Meanwhile, the IIF coincident index, denoted by “IIF-CI”, follows the same approach as the BdL-CI but includes an additional five variables Iradian & Zouk (2010). ^[4]

While providing a useful gauge of economic activity in Lebanon, our analysis reveals that the statistical properties of these two coincident indexes – such as accuracy and unbiasedness – could be improved. As illustrated Figure 1, the IIF-CI, consistently underestimated actual GDP growth between 2006 and 2011. On the other hand, the BdL-CI had no systematic estimation error; however, its accuracy has recently been weak as the gap between the BdL-CI and actual growth has been relatively large in several years, and in one year (2006), the BdL-CI also qualitatively misdiagnosed the strength of economic activity (the BdL-CI signaled that the economy was contracting by 1.4 % while it actually grew by 1.6 %).

Figure 1:

Yearly growth rates of BDL-CI, IIF-CI and real GDP.

timeliness and accuracy of economic activity estimates in Lebanon, we design two new indexes for the Lebanese economy: a coincident (WB-CI) and a leading (WB-LI) index. This study builds upon features from the empirical literature on coincident and leading indices such as Moore and Shiskin (1967), Stock and Watson (1989), Chauvet (1998), Mariano and Murasawa (2002), Issler and Vahid (2006), Pereira (2012) and Issler et al. (2013).

In this paper we add to the literature two main contributions. First, we improve on the NBER-Conference Board approach (Conference Board 2014) by employing a minimization technique to ameliorate the statistical properties of coincident and leading indexes. We do so by calibrating the weights assigned to each variable so that the yearly growth rate of the constructed index (Coincident or Leading) converges to the actual growth rate of the reference variable (in our case GDP) at any point in time. Second, we develop for the first time a leading index for the Lebanese economy that can be used to (i) detect early signs of turning points in economic activity and (ii) forecast GDP growth during the next 12 months.

The rest of this paper is structured as follows. Section 2 presents a brief historical overview of coincident and leading indexes. Section 3 describes the methodologies used to construct the WB-CI and WB-LI. Section 4 presents the results while Section 5 discusses some policy implications derived from these newly developed indexes. Finally, Section 6 presents the caveats and limitations of the indexes and Section 7 concludes.

2 Literature Review

Knowing the current and the expected state of the economy is of foremost importance for businesses, policymakers and investors. Depending on the quality of statistics in a given country, two distinct methodologies have been used to develop coincident and leading indexes: (i) the National Bureau of Economic Research (NBER) and/or Conference Board (CB) methodology and (ii) the Stock and Watson (1989) methodology (hereafter denoted by SW). The NBER-CB approach is a non-econometric technique, while the SW methodology is based on advanced statistical properties such as Dynamic Factor and Markov Switching models.

The idea of quantitatively monitoring a business cycle, which “commonly refers to co-movements in different forms of activity, not just fluctuations in GNP” (Stock and Watson 1989, 353), was first introduced in 1938 by a research team at the NBER. This team, which was led by Wesley Mitchell and Arthur Burns, examined the dynamics of some economic variables to see whether these changes lagged, led or coincided with changes observed in US business cycles. Twenty years later, Moore and Shiskin (1967) developed, for the first time and based on Burns and Mitchell’s (1946) research, a methodology to construct composite indexes of real economic activity. They designed “a scoring plan that has been developed to help in the evaluation and selection of indicators” (Moore and Shiskin 1967, 3). The scoring plan of each variable was based on (i) statistical adequacy, (ii) timeliness of publication, (iii) smoothness, (iv) economic significance, (v) historical business cycle conformity, and (vi) cyclical timing. This approach, which was adopted by the Conference Board in 1995 and developed later on, is, however, subject to some criticism. First, many econometricians argued that the selected variables were subjectively chosen without any econometric foundation (Marcellino 2005). This is what Koopmans (1947) described as a “measurement without theory”. Second, and as argued by Chauvet (1998) indexes constructed with this methodology are constantly revised ex-post and since revisions can be significant in certain cases, real-time assessment of the business cycle may be severely impacted.

In a first attempt to respond to these criticisms, Stock and Watson (1989) developed an econometric model to construct new coincident and leading indexes for the U.S. economy. In their method, the coincident indicator (or Index) is represented by an unobserved reference cycle representing what they referred to as the “state of the economy”. The index is then measured using a Dynamic Factor model, where the parameters of the series ^[5] forming the index are estimated using Maximum Likelihood and Kalman Filter methods. In addition, they developed a leading indicator that forecasts the growth rate of the coincident index over the next six months, using a set of variables ^[6] in a Vector Autoregressive model (VAR). However, the SW approach failed to anticipate the U.S. recession in 1990–1991 because it assumed that expansions and contractions are symmetric in terms of their impact on the economy. For that reason, and in order to account for the fact that recessions are more sudden and shorter than expansions, Chauvet (1998) improved on the SW methodology by capturing nonlinearities in the business cycle using the Markov Switching model as proposed by Hamilton (1989). Other noticeable scholars that tried to improve on the SW econometric methodology are Mariano and Murasawa (2002) who employed a mixed frequency model by adding quarterly real GDP data to the initial set of monthly indicators and Issler and Vahid (2006) who used canonical correlation analysis and an instrumental variable probit model to construct a coincident index of the US economy.

Historically, economic research in this area was focused on developed countries. However, as a result of the improvement in the quality and frequency of data, research examining business cycles in emerging countries increased in recent years. Consequently, economists deepened their understanding of the business cycle in many of these countries by developing coincident and/or leading indexes. For example, Saadi-Sedik and Mongardini (2003) employed a reduced form model of the SW methodology to construct coincident and leading indexes for the Jordanian economy from January 1996 to December 2002. More recently, Pereira (2012) constructed a quarterly coincident index for the Cape Verdean economy, from 2002 to 2010, using the NBER/CB approach, while Issler et al. (2013) designed coincident indexes for Argentina, Brazil, Chile, Colombia and Mexico. Finally, Dorsainvil (2014) designed a coincident index for the Haitian economy using the Principal Component Analysis approach.

Although the SW methodology provides a proper statistical framework, it can’t be applied in the Lebanese context due to the absence of long historical time-series. In particular, we only have yearly data for the Lebanese GDP from 2004 to 2011 which only gives us eight observations. Therefore, the law of large numbers, on which the relevant time-series econometric techniques depend, will be violated, hence decreasing the power of these models (higher probability of committing a Type II error). For that reason, we will use an improved NBER-CB approach which can be applied in countries that suffer from weak statistical systems (similar to Lebanon). This approach has the important advantage of being easy to construct, explain and analyze. These are valuable properties in constructing composite indexes aimed at a wide public.

3 Methodology

4 Results of the WB-CI and the WB-LI

Our results indicate that the designed WB-CI measures the economic activity in Lebanon very accurately. ^[15] The criteria used to assess the effectiveness of the WB-CI in monitoring the Lebanese economic activity is based on calculating the absolute value of the error between the yearly growth rate of the WB-CI and the actual (realized) annual GDP growth rate. When this error tends, on average, to 0 and its standard deviation converges to 0 this implies that the WB-CI accurately reflects the dynamics of the Lebanese real economy. As reported in Table 5, the average error and the standard deviation between 2006 and 2011 were both equal to 0.0 %. This compares favorably to both the BdL and the IIF coincident indexes. For the BdL-CI (IIF-CI), the average growth rate was 2.37 (1.53) % and the standard deviation was 1.29 (1.14) % during the same period. Consequently, the WB-CI is a reliable measure of the current state of the Lebanese economy.

The one-year ahead forecasting performance of the constructed WB-LI is encouraging. ^[16] As illustrated in Figure 2, the yearly average growth rate of the WB-LI leads the yearly growth rate of the WB-CI by a year. Furthermore, Figure 3 shows that the yearly growth rate of the leading indicator at time t−12 follows almost the same pattern as the yearly growth rate of the WB-CI at time t, suggesting that the WB-LI should be a useful forecasting tool. In fact, between December 2008 and December 2013, the average error between the yearly growth rate of the WB-LI at time t−12 and the yearly growth rate of the WB-CI at time t was only 0.37 %, while the standard deviation of this error was 0.28 %.

Figure 2:

The WB-LI is a good tool to detect turning points in the economy….

Figure 3:

… and forecast economic growth in the short-run.

To illustrate the importance of having timely and accurate economic data, it is instructive to look at the monetary policy implications of the (new) WB-CI versus that of the existing BdL-CI. As detailed below, the latest reading from the BdL-CI points to a constant growth in economic activity. The WB-CI, however, is showing an economy that is accelerating. The growth rate of the BdL-CI was constant at 3.2 % between 2013 and 2014, suggesting that the Lebanese economy grew at the same pace. In contrast, the WB-CI showed that the economy grew by 1.2 % in 2014 compared to 0.8 % during 2013, reflecting an acceleration in economic activity. A sustained period with such diverging trends would result in sharply different monetary policy decisions.

5 Some Policy Implications

In this section, we examine the impact of unexpected changes in macroeconomic variables on economic growth using the WB-CI decomposition. We employ a Vector Autoregressive (VAR) model to examine the relationship between the variables forming the WB-CI and real GDP growth as proxied by the growth rate of the WB-CI itself. A necessary condition for VAR models to be econometrically valid is the stationarity of variables. Table 10, in the Appendix, shows that the log differences of WB-CI, tobacco excise (TE), cement deliveries (CD), VAT revenues (VAT), tourist arrivals (TA), taxes on real estate (TR), money supply (M3), lending to the private sector (LP), administrative fees and charges (AC), non-resident deposits (NRD), primary spending (PS), and cleared checks (CC) are found to be stationary at the 10 % level. ^[17] Let ΔWB−CIt and Δyk;t denote the one period growth rates of the WB-CI and variable kk at time (or month) $t=Dec−2004,…,Oct−2013, respectively. The VAR(k) model is given by:

where Xt,k=[ΔWB−CIt,Δyk;t] is the vector of endogenous variables and εt,k denotes the error vector of the VAR(k.) model. In addition, Lk denotes the optimal lag length of the VAR(k) model as determined using the Akaike Information Criterion (AIC).

We proceed to generate, the Impulse Response Functions (IRFs) for the VAR(k) model. The results of the IRFs, illustrated in Figure A1 (in the Appendix), trace the response of real economic activity (proxied by the WB-CI) to one standard deviation shocks in each variable k over a 24 months period. We only comment on the IRFs that are statistically significant. ^[18] In particular, the WB-CI reacts positively to a shock in M3 and CD suggesting that the monetary policy (conducted by BdL though M3) and the real-estate sector are major drivers of growth in Lebanon. As such, BdL is encouraged to maintain its expansionary monetary policy aimed at boosting economic activity. Moreover, higher business activity, proxied by a positive shock in CC, enhances growth. Therefore, improving the political and security conditions, which tend to positively affect consumers’ and investors’ sentiment, as well as improving the business environment are key to stimulate growth. Finally, the IRFs of WB-CI to shocks in TA, TE and VAT are also positive. To show the relative importance of the (statistically significant) shocks to the variation of the WB-CI, we generate the Forecast Error Variance Decomposition (FEVD). The FEVD analysis, illustrated in Table 11, reveal that all the k′k′ variables explain a large part of the variation in real economic activity in the short and long run.

In order to examine causality between the k′ variables and the real economic activity, we apply Granger Causality tests to the VAR(k′) model of [eq. 8]. ^[19] As shown in Table 12, the results show that there is a bi-directional Granger causality between WB-CI and each of tobacco excise, cleared checks, cement deliveries, administrative fees and charges and VAT revenues over current real economy. On the other hand, there is a unidirectional causality from with WB-CI Granger causing M3.

Finally, and based on the significant IRFs and the Granger causality tests, the economy would be expected to grow, as show in Table 13, by 0.03, 0.2, 0.21, 0.21 and 0.32 percentage points next year if we observe a one percentage point increase in cement deliveries, cleared checks, tobacco excise, administrative fees and charges and VAT revenues respectively this year.

6 Caveats and Limitations

While the newly developed indexes reveal promising statistical properties in terms of estimating and forecasting real economic activity in Lebanon, analysts and policy makers should be aware of the following limitations of the model when interpreting the results:

1. First, not all the sectors of the economy were adequately represented in the pool of potential variables due to data limitations. Most notably, good proxy variables were not available to accurately account for the agriculture and industry sectors, which combined accounted for 18.6 % of GDP in 2011.

2. Second, due to data limitations we deflated all the monetary variables into real terms using the CRI consumer price index that only account for the price dynamics in the greater capital area (greater Beirut). However, the Central Administration of Statistics (CAS) has recently (March, 2014) rebased Lebanon’s inflation data to December 2013 and now provides (i) a much more comprehensive breakdown of prices (prices for rent are now collected on a monthly basis) and (ii) an updated weighting scheme to the inflation basket. As a result, this improved CAS CPI series could be preferable to the one currently used, once the old (2007) CPI index is rebased to the new (2014) index in order to have a continuous and comparable CPI series.

3. Third, the lag period of the WB-CI and the WB-LI appeared to be larger (4–5 months) than initially expected due to the infrequent and lagged release of the fiscal data from the ministry of finance.

4. Fourth, and as illustrated in figure 4, economic activity in Lebanon is highly volatile and subject to frequent exogenous shocks of political and security nature. However, this volatility is not captured by the WB-LI as it is hard to predict sudden changes in the political and/or security conditions. As a result, analysts should take into account any recent political developments rather than interpreting the results mechanically.

5. Fifth and most importantly, the yearly growth rate of the newly constructed indexes were only tested against six actual GDP growth observations (2006–2011) because longer time series were not available. As a result, some non-negligible margin of error is expected to remain in the estimates and results. In order, to improve the efficiency of the estimates, the model will be recalibrated whenever new national accounts data are published by CAS.

Figure 4:

Estimated quarterly growth based on WB-CI and WB-LI (yoy).

7 Conclusion

This paper presents a new coincident index (WB-CI) and a 12-months leading index (WB-LI) for the Lebanese economy. These indexes, which can be used as monthly proxies for the evolution of real GDP, were constructed using an improved methodology based on the NBER-CB approach.

Notwithstanding the relatively small sample period, the results reveal promising statistical properties that should make these new indications valuable coincident and leading (one-year ahead) indexes for monitoring the Lebanese economy. However, given limitations on the length of time series in Lebanon, the accuracy and reliability of these indexes in tracking the business cycle of the Lebanese economy is expected to improve over time as more data points become available. To take into account the statistical value of these new data points, both indexes will be re-estimated periodically. It is therefore likely that the composition of the WB-CI and the WB-LI will change in the coming years and become more stable as actual GDP series from national authorities are published. Furthermore, and rather than analyzing the results of these indexes mechanically, analysts should incorporate into their assessment of economic activity historical, political and cultural factors that do not easily lend themselves to quantification.

Finally, this new approach of adding a minimization problem to the original NEBR-CB methodology may provide a useful roadmap to analyze business cycles in countries that have weak economic statistics.