Constructing an Economic Activity Indicator for Turkey

Transcription

1. Constructing an Economic Activity Indicator for Turkey Aysu Çelgin Elif Akbostancı July 2021 Working Paper No: 21/14
2. © Central Bank of the Republic of Turkey 2021 Address: Central Bank of the Republic of Turkey Head Office Structural Economic Research Department Hacı Bayram Mah. İstiklal Caddesi No: 10 Ulus, 06050 Ankara, Turkey Phone: +90 312 507 80 04 Facsimile: +90 312 507 78 96 The views expressed in this working paper are those of the author(s) and do not necessarily represent the official views of the Central Bank of the Republic of Turkey.
3. Constructing an Economic Activity Indicator for Turkey # Aysu Çelgin*, Elif Akbostancı** Abstract In this paper, a monthly economic activity indicator is constructed for the Turkish economy for 1988-2020 period. Dynamic factor modelling framework is utilized in the estimation of the indicator. In the context of data selection, first of all, the variables are categorized into five types as: activity (hard data), activity (survey-based data or soft data), trade, employment and financial variables. After determining the candidate variables for each category, data selection is finalized by using the hard-thresholding method. The results indicate that monthly economic activity indicator is successful in detecting the past recessionary and contractionary periods of the Turkish economy and providing timely information about the course of the economic activity. Keywords: Economic activity, Dynamic factor model, Hard-thresholding, Real time analysis JEL Classification: C22, E32, E37 * Corresponding author. Central Bank of the Republic of Turkey, Research and Monetary Policy Department. ** Middle East Technical University, Department of Economics. # We would like to thank Mahmut Günay, Çağlar Yüncüler and Çağrı Sarıkaya for their invaluable contributions and comments. The views expressed in this paper are those of the authors and do not reflect the official views of the Central Bank of the Republic of Turkey.
4. Non-Technical Summary In this paper , a monthly economic activity indicator for the Turkish economy for 1988-2020 period is constructed. Dynamic factor modelling framework is utilized in the estimation of the indicator. This indicator can be used to assess the expansion/contraction phases of the economic activity. In the variable selection part, we implement the hard-thresholding method by Bai and Ng (2008). We firstly categorize the variables into five types as: activity (hard data), activity (survey-based data or soft data), trade, employment and financial variables. After doing the categorization, candidate variables for each category are determined. Then by using the hardthresholding method, the most parsimonious variables are obtained for estimation of the economic activity indicator from the available data set. Based on this methodology, we construct an economic activity indicator using GDP, industrial production index, electricity production, total vehicles production, volume of production over the past 3 months, real sector confidence index, import volume index, non-farm employment, credit stock and credit default swap (CDS). Our monthly economic activity indicator detects October 1988-February 1989, April 1994January 1995, October 1998-March 1999, August 1999-September 1999, February 2001February 2002, October 2008-September 2009 and August 2018-January 2019 as recessionary/contractionary periods. A real-time application of the model confirms that using timelier variables provide timely information about the current state of the economy.
5. 1 . Introduction and Literature Review Economic activity is one of the main indicators to evaluate how a country performs over time. Academicians, business people and decision makers give much importance to the detection of expansion/contraction phases and the real-time cyclical analysis of the economic activity as it is considered crucial for implementing efficient policies. To this end, accurate assessment of the economic activity has an important place in the economic literature. Gross Domestic Product (GDP) is the most widely known indicator for making deductions about the overall performance of economic activity. However, it has certain drawbacks regarding its timing and content. First of all, it is published with a considerable time lag and is exposed to data revisions after the initial dissemination. Second, GDP itself is not enough to encompass all information about the overall economic activity (Aruoba and Sarikaya, 2013). Tracking the developments in markets other than goods market, e.g. labor market, financial markets, etc., necessitates different data sources and more information than GDP provides. Moreover, utilizing soft data (survey-based data) related with production, consumption, trade and employment may also be useful in providing timelier information and incorporating expectations (Aruoba, Diebold and Scotti, 2009). To sum up, a timelier indicator of economic activity, which encompasses different branches of the economy, would be preferable to GDP. The literature on constructing economic activity indicators is well established, in particular for the US economy. It has progressed over time in terms of data type and estimation methodologies. The pioneering study of Stock and Watson (1989) presents three monthly economic activity indicators for the US economy by using different macroeconomic variables on monthly frequency with the help of dynamic factor modelling framework. Other examples for the US economy that prefer dynamic factor modelling as estimation framework are Mariano and Murasawa (2003), Proietti and Moauro (2006) and Auroba et al. (2009). These papers differ from Stock and Watson (1989) in selecting variables at different frequencies, such as quarterly, weekly and daily, to their data set. They criticize Stock and Watson (1989) in exclusion of quarterly indicators, especially GDP, since this prevents utilization of extra information in quarterly variables. They extend the work of Stock and Watson (1989) by incorporating GDP in their data set, as of being the most important measure of the overall state of the economy. In a later study, Evans (2005) contributes to the literature by estimating the state of the US economy in real-time on a daily basis. Furthermore, Aruoba et al. (2009) construct an indicator on a daily basis for the US economy. Proietti and Moauro (2006) extend their work by
6. constructing economic indicators for Euro Area . In a later paper, Matheson (2011) also extends the country coverage to 32 countries by utilizing dynamic factor modelling approach and monthly variables out of 6 different variable blocks in the construction of economic activity indicator. Darne and Ferrara (2011) construct indicators for both Euro Area and six main countries by using a Markov-Switching VAR model and Markov-Switching factor model. Additionally, Dua and Banerji (2000) and Simone (2001) construct economic activity indicators by using NBER methodology and general to specific approach for Indian economy and Argentina, respectively. The literature on constructing economic activity indicator for Turkey is scarce. Among them, Aruoba and Sarikaya (2013) construct an economic activity indicator by using mixed frequency variables. Additionally, Çakmaklı and Altuğ (2014) construct a coincident real economic activity indicator for Turkey. Both studies use dynamic factor modelling, while the latter utilize Bayesian semiparametric estimation instead of Kalman filtering algorithm used by the former. In this paper, we construct a monthly economic activity indicator to detect historical expansion/contraction periods and to evaluate the economic outlook of the Turkish economy so that we provide timely information about the near future of economic conditions in Turkey. Similar to most of the studies in the literature, dynamic factor modelling approach is preferred in constructing economic activity indicator because of being good at synthesizing macroeconomic variables into an indicator (Barhoumi, Darne and Ferrara, 2013). Based on this methodology, we construct an economic activity indicator using the industrial production index, electricity production, total vehicles production, volume of production over the past 3 months, real sector confidence index, import volume index, non-farm employment, credit stock and credit default swap (CDS) in the estimation. GDP is also added to the data set since it is the broadest measure of economic activity. One of the main contributions of this study is that the variables are selected in an analytical and systematic way. First of all, the variables that are used in the analysis are grouped in five different categories which are activity (hard data), activity (survey-based data or soft data), trade, employment and financial variables. After doing the categorization, candidate variables for each category are determined. In the construction of our data set for the economic activity indicator, we implement the hard-thresholding method by Bai and Ng (2008). By this method, the most parsimonious variables are obtained for estimation of the economic activity indicator from the available data set.
7. Another contribution of this paper is that we use a timelier data set compared to other two papers about Turkey , i.e. Aruoba and Sarikaya (2013) and Çakmaklı and Altuğ (2014). Nearly all variables used in these two papers are released with a lag of 40 to 60 days. Although our data set also contains variables announced with lagged periods, six of the ten variables are released at the appertaining month or at the beginning of next month. In this respect, we can update our indicator earlier than the other two studies and get timelier signals about the course of economic activity. The last contribution of our work is that our indicator is good at detecting the past recession and contraction periods. Aruoba and Sarikaya (2013) and Çakmaklı and Altuğ (2014) construct indicators for 1987-2011 and 1989-2014 periods respectively and they detect nearly the same periods, which are 1994, 2001 and 2008-2009, as recession periods for Turkish economy. The monthly economic activity indicator constructed in our paper covers the period starting from 1988 to February 2020. In our work, we implement the same approach with these two papers when determining the recessionary periods. We detect that October 1988-February 1989, April 1994-January 1995, October 1998-March 1999, August 1999-September 1999, February 2001-February 2002, October 2008-September 2009 and August 2018-January 2019 are recessionary and contractionary periods. This means that our indicator detected both the recessions indicated in these two papers and four more periods as recessions and contractions which are October 1988-February 1989, October 1998-March 1999, August 1999-September 1999 and August 2018-January 2019. The results show that the economic activity indicator is good at detecting historical recession and contraction periods. To evaluate whether the indicator provide timely information about the current state of the economy or not and to show the importance of using timelier variables in the estimation, a real-time application is performed. In this regard, the model is estimated until a certain period and the parameters are fixed; then, the economic activity indicator is calculated at new data announcements for different data releases. It is concluded that with the timelier variables announced more promptly, we would get timely information about declines in economic activity. For this study, we are able to assess the decline in economic activity starting from April 2020, even in that month. The paper proceeds as follows. Section 2 provides the model and methodology used in the estimation of the indicator. Section 3 introduces the data set and variable selection methodology in detail. Section 4 presents the economic activity indicator. Additionally, implied recession and contraction periods indicated for Turkish economy are compared with other
8. papers on Turkey and elaborated in detail, and also a real-time application for the Turkish economy is presented. Finally, Section 5 presents a brief summary of the findings and concludes the paper. 2. Model and Methodology As for the use of dynamic factor modelling (DFM) framework in the construction of economic activity indicators, Barhoumi et al. (2013) state that it is a useful tool in summarizing the information in a large data set by forming a small number of common factors. In other words, DFM framework is generally opted for its success in utilizing the macroeconomic variables and producing reliable estimates. In the construction of our economic activity indicator, we follow a similar approach to Aruoba and Sarikaya (2013). In this method, economic conditions are considered as unobserved variable and tried to be explained by different observed indicators. Furthermore, variables in different frequencies can be used simultaneously. Then, with the help of a linear and statistically optimal filter, the economic activity indicator is calculated. 2.1. Dynamic Factor Model at Monthly Frequency Although economic conditions change at higher frequencies (hourly, daily, etc.), data releases have been less frequent. Most of the indicators are often monthly or quarterly. Therefore, we construct our indicator at monthly frequency. If the data has a higher frequency than monthly frequency, two different approaches are taken depending on the characteristics of the data. If it is a stock variable, end of the month value is taken as the monthly data. If it is a flow variable, then monthly average of the data is calculated. As mentioned earlier, we adapt the methodology of Aruoba and Sarikaya (2013) while constructing the model. In this framework, the unobserved economic conditions at month t is denoted by