QUEST_Serbia DSGE Model and Data

Transcription

1 Miroljub Labus QUEST_Serbia DSGE Model and Data v Belgrade December

2 Agenda 1. Data 2. Model Calibration and Estimation 2

3 Part 1. Data update for ESA standard 3

4 Part 1. Content All time series are updated for the period from the first quarter of 2003 to the third quarter of 2014, and adjusted to the new ESA standard (European System of Accounts) : 1. New ESA 2010 standard, 2. Data collection, 3. Data processing, 4. Original and deseasonalized data, 5. Statistics. 4

5 Part 1.1. ESA 2010 standard Revised data release The Statistical Office of Serbia (SOS) released on November 28, 2014 the latest update of quarterly data set for Serbia s GDP, SOS has adopted the new methodology for compiling national accounts, which is known as ESA 2010 standard European System of Accounts, ESA 2010, Eurostat and European Commission 2013, The new European System of National and Regional Accounts (ESA 2010) is a major development of the previous version of 1995, It was released in 2013, and adopted by SOS in 2014, GDP are re-estimated according to the new standards, by final demand components and by sectors of value-added origin, for the period from Q1Y2003 to Q3Y2014, The new methodology adjusted upward GDP series for an average of 5.8%, and made different estimations for some GDP components, This new data set forms the empirical background for the current version of QUEST_Serbia DSGE model. 5

6 Part 1.2. Data source The source data file: "Kvartalni BDP - Q12001 Q slanje" Real data: Sheet "BDP ulancane mere obima ref2010" Real consumption: gdp_c = col(ac+ad) Real public consumption: gdp_g= col(ae) Real investment: gdp_i_org= col(af) Real export: gdp_ex = col(ag) Real import: gdp_im = col(ah) Real GDP: gdp = col(ai) Nominal data: Sheet "BDP tekuce cene" Nominal consumption: gdp_c_tc = col(ac+ad) Nominal public consumption: gdp_g_tc = col(ae) Nominal investment: gdp_i_tc_org = col(af) Nominal export: gdp_ex_tc = col(ag) Nominal import: gdp_im_tc = col(ah) Nominal GDP: gdp_tc = col(ai) 6

7 Part 1.3. Data processing Data processing procedure is similar to the one adopted in the previous release of the model; however, for the sake of completeness we will repeat it at this place, There is always a residual component in GDP due to errors and omissions in complying GDP data. The residual component of GDP is implicitly defined as a difference between the value of GDP at constant prices and the sum of GDP components. The Government investments are missing from statistical data on GDP. Government investments are therefore approximated by the series of fiscal capital expenditures; this data are provided by the Ministry of Finance. Nominal Government investments are deflated by the implicit GDP deflator of investment goods. Private investment component is given by a difference between total investments and government investments. Private investments are simple called investments, while public sector investment are called government investments. 7

8 Part 1.3. Data processing CONTINUE GDP deflator: Deflators defl_gdp = gdp_tc/gdp Consumption deflator: defl_c = gdp_c_tc/gdp_c Public consumption deflator: defl_g = gdp_g_tc/gdp_g Export deflator: Import deflator: Investment deflator: defl_ex = gdp_ex_tc/gdp_ex defl_im = gdp_im_tc/gdp_im defl_i = gdp_i_tc_org/gdp_i_org Nominal public investment: gdp_gi_tc >>> fiscal data, table 3, capital expenditures Real public investment: gdp_gi = gdp_gi_tc/defl_i 8

9 Part 1.3. Data processing CONTINUE GDP and its components are deseasonalized by X13 EViews procedure (using logarithm transformation of data). Deseasonalized GDP and the GDP obtained as a sum of its deseasonalized components has a small difference. That residual is assigned to the investment series. GDP of the Euro zone (18 members) is taken by the market prices from the ECB website. This nominal GDP is deflated by the series of HICP with 2010=1. Real EU GDP is deseasonalized by X13 EViews procedure. The QUEST_Serbia_data.m file will normalized GDP Serbia to GDP EU to one by using an appropriate scaling factor. The EU price series is normalized to 2010=1. Domestic and foreign interest rates (inom and inomw) are written in decimal points. QUEST_Serbia_data.m will make them as quarterly rates by dividing the corresponding data by 4. 9

10 Part 1.3. Data processing CONTINUE The nominal exchange rate is taken from the NBS website and normalized to one for the 2010 average. The starting point for the gross wage bill is taken from statistics reported in the file Breakdown of wages by budget and property types internally released by the Statistical Office to the Ministry of Finance. This statistics takes into account only 1.7 million people and only taxes and contributions paid by employees. There are two missing parts in this statistics. The first missing part is fixed by including non-civilian workforce (police and military). The second missing part is fixed by multiplying taxes and contributions by the factor 1,179 in order to encompass employers taxes and contributions. 10

11 Part 1.3. Data processing CONTINUE Finally, the obtained figures are checked by comparing them, year by year, with figures on employees compensation released by the Statistical Yearbook s annual data on GDP by the income type. The gross wage bill obtained in this way is dividing by the enhanced number of employees to get the average gross wage rate. This gross wage rate is afterwards deseasonalized by using X13 EViews procedure. The same was done for the enhanced number of employees. 11

12 Parts 1.4. Original and Deseasonalized Time Series GDP, Consumption, Government consumption, Private investment, Government investment, Export and Import, Original series are displayed in solid blue lines, Deseasonalized series are displayed with red dots. 12

13 Part 1.4. Updated GDP for 2014 Old methodology New methodology 800, , , , , ,000 Deseasonalized GDP 840, , , , , ,000 Deseasonalized GDP 560, , , , , ,

14 Part 1.4. Updated Consumption for 2014 Deseasonalized consumption 640, , , , , , ,

15 Part 1.4. Updated Government Consumption for 2014 Deseasonalized government consumption 170, , , , , , ,

16 Part 1.4. Updated Private Investments for 2014 Deseasonalized private investment 240, , , ,000 80,000 40,

17 Part 1.4. Updated Government Investments for 2014 Deseasonalized government investment 70,000 60,000 50,000 40,000 30,000 20,000 10,

18 Part 1.4. Updated Export for 2014 Deseasonalized export 400, , , , , , ,

19 Part 1.4. Updated Import for 2014 Deseasonalized import 500, , , , , , ,

20 Parts 1.5. Statistics for Deseasonalized Time Series GDP, Consumption, Government consumption, Private investment, Government investment, Export and Import. 20

21 Part 1.5. Statistics for GDP Old methodology New methodology Series: GDP Sample 2003Q1 2014Q1 Observations 45 Mean Median Maximum Minimum Std. Dev Skewness Kurtosis Jarque-Bera Probability Series: GDP Sample 2003Q1 2014Q3 Observations 47 Mean Median Maximum Minimum Std. Dev Skewness Kurtosis Jarque-Bera Probability GDP estimated according to the previous methodology had 12% probability to fit the normal distribution pattern, GDP estimated according to the new methodology has 7% probability to be normally distributed, However, null hypotheses that deseasonalized GDP is normally distributed must be completely rejected Series: GDP_SA Sample 2003Q1 2014Q3 Observations 47 Mean Median Maximum Minimum Std. Dev Skewness Kurtosis Jarque-Bera Probability

22 Part 1.5. Statistics for GDP The annualized rates of growth of GDP are the same whatever the methodology for estimating national accounts were used, Those rates are not normally distributed GROWTH_GDP_PCA Series: GROWTH_GDP_PCA Sample 2003Q1 2014Q3 Observations 46 Mean Median Maximum Minimum Std. Dev Skewness Kurtosis Jarque-Bera Probability

23 Part 1.5. Statistics for Consumption Series: GDP_C_SA Sample 2003Q1 2014Q3 Observations 47 Mean Median Maximum Minimum Std. Dev Skewness Kurtosis Jarque-Bera Probability

24 Part 1.5. Statistics for Government Consumption Series: GDP_G_SA Sample 2003Q1 2014Q3 Observations 47 Mean Median Maximum Minimum Std. Dev Skewness Kurtosis Jarque-Bera Probability

25 Part 1.5. Statistics for Private Investments Series: GDP_I_GI_SA Sample 2003Q1 2014Q3 Observations 47 Mean Median Maximum Minimum Std. Dev Skewness Kurtosis Jarque-Bera Probability

26 Part 1.5. Statistics for Government Investments Series: GDP_GI_SA Sample 2003Q1 2014Q3 Observations 47 Mean Median Maximum Minimum Std. Dev Skewness Kurtosis Jarque-Bera Probability

27 Part 1.5. Statistics for Export Series: GDP_EX_SA Sample 2003Q1 2014Q3 Observations 47 Mean Median Maximum Minimum Std. Dev Skewness Kurtosis Jarque-Bera Probability

28 Part 1.5. Statistics for Import Series: GDP_IM_SA Sample 2003Q1 2014Q3 Observations 47 Mean Median Maximum Minimum Std. Dev Skewness Kurtosis Jarque-Bera Probability

29 Part 2. Model Calibration and Estimation 29

30 Part 2. Contents 1. Model initialization, 2. Bayesian estimation of shocks and parameters, 3. Model s updating time series, 4. In-sample forecast, 5. Simulation. 30

31 Part 2.1. Model Initialization 31

32 Part 2.2. Bayesian Estimation Bayesian estimation is somewhere in between calibration and maximum likelihood estimation, The tradition of calibrating models is inherited through the specification of priors, The maximum likelihood approach enters through the estimation process based on confronting the model with data, Priors are plotted by gray density functions, Posteriors are plotted by solid density functions, and The vertical green line is the kernel, The closer the maximum or saddle point of the posterior density function to the kernel, the better the estimation of the underlying parameter. 32

33 Part 2.2. Priors and Posteriors of Shocks 33

34 Part 2.2. Priors and Posteriors of Shocks CONTINUE 34

35 Part 2.2. Priors and Posteriors of Parameters 35

36 Part 2.2. Priors and Posteriors of Parameters CONTINUE 36

37 Part 2.2. Priors and Posteriors of Parameters CONTINUE 37

38 Part 2.2. Priors and Posteriors of Parameters CONTINUE 38

39 Part 2.2. Priors and Posteriors of Parameters CONTINUE 39

40 Part 2.3. Model s Updated Series The model computes the expected value of endogenous variables given the information available at the current date (E t y t ), Updated time series are displayed in green dashed lines, Empirical time series are displayed in solid blue lines, The model is able to correctly replicate most of the empirically based macroeconomic variables, The time span of empirical variables is truncated to the period between the first quarter of 2007 and the first quarter of

41 Part 2.3. Updating Series 41

42 Part 2.3. Updating Series CONTINUE 42

43 Part 2.3. Updating Series CONTINUE 43

44 Part 2.3. Updating Series CONTINUE 44

45 Part 2.3. Updating Series CONTINUE 45

46 Part 2.4. In-Sample Forecast The model computes the posterior distribution of filtered endogenous variables or one-step ahead forecasts (E t y t+1 ), It demonstrates the model ability to do in-sample forecast of endogenous variables, The forecasting is based on all available information at the present time, Green lines represent deciles of the forecasted variables, The solid black line is the mean forecast, Time series are truncated to 29 data points representing endogenous variables in the period between the first quarter of2007 and the first quarter of

47 Part 2.4. One-Step Ahead Forecast 47

48 Part 2.4. One-Step Ahead Forecast CONTINUE 48

49 Part 2.4. One-Step Ahead Forecast CONTINUE 49

50 Part 2.4. One-Step Ahead Forecast CONTINUE 50

51 Part 2.4. One-Step Ahead Forecast CONTINUE 51

52 Part 2.5. Simulation The model computes impulse response functions taking into account all information from the beginning of time horizon up to the first quarter of 2014, The model computes marginal and cumulative IRFs, The model, also, takes decomposition of effects generated by the state variables on other endogenous variables, By this way, it is provided additional incites into general equilibrium dynamics between macroeconomic variables, We will illustrates this by focusing only on impacts of an increase of the repo interest rate on output, real interest rate, inflation and real exchange rate. 52

53 Part 2.5. Monetary Policy: Marginal IRFs 53

54 Part 2.5. Decomposed Output Growth Rate s IRF 54

55 Part 2.5. Decomposed Inflation Rate s IRF 55

56 Part 2.5. Decomposed Real Exchange Rate s IRF 56

57 Part 2.5. Decomposed Real Interest Rate s IRF 57