This action might not be possible to undo. Are you sure you want to continue?
Blominvest Working Paper Research Department
Estimating Real GDP Growth for Lebanon
Prepared by Marwan Mikhael, Michel G. Kamel, and Gaelle Khoury Authorized for Distribution by Dr. Fadi Osseiran December 2010
ABSTRACT This paper presents an econometric framework to estimate quarterly GDP growth for Lebanon based on a bottom up approach from the demand side. The model relies on a data set of 68 quarterly observations from 1993 to 2009 for ten endogenous variables and two exogenous variables selected on the basis of their economic and statistical significance. Quarterly GDP figures are obtained from the annual series using the Chow-Lin disaggregation method. The model derived is a Vector Autoregressive Model with Exogenous Variables (VARX), a variant of the Vector Autoregressive Model (VAR) that takes into account both exogenous and endogenous variables. Our empirical results show robust correlation between the estimate and actual quarterly GDP figures, indicating the ability of the model to provide a high level of accuracy in estimating real GDP growth. JEL Classification Numbers: C51, C32, C13, C82, O53 Keywords: Granger causality, VAR, VARX, impulse response function, and GDP estimate.
Disclaimer: This report is published for information purposes only. The information herein has been compiled from, or based upon sources we believe to be reliable, but we do not guarantee or accept responsibility for its completeness or accuracy. This document should not be construed as a solicitation to take part in any investment, or as constituting any representation or warranty on our part. The consequences of any action taken on the basis of information contained herein are solely the responsibility of the recipient. Copyright 2010 BLOMINVEST SAL No part of this material may be copied, photocopied or duplicated in any form by any means or redistributed without the prior written consent of BLOMINVEST SAL. 1
Table of Contents
I- Introduction.................................................................................................................................. 3 II- Literature Review ........................................................................................................................ 5 III- GDP Disaggregation in the Case of Lebanon .............................................................................. 6 IV- Choice and Characteristics of Variables ..................................................................................... 8 V- The Vector Autoregressive Model (VAR) .................................................................................. 10 VI- Vector Autoregressive with Exogenous Variables (VARX) ....................................................... 14 VII- Impulse response .................................................................................................................... 18 VIII- Conclusion .............................................................................................................................. 21 References ..................................................................................................................................... 22 Appendix A - VAR Stationarity ....................................................................................................... 24 Appendix B - Vector Autoregressive Model (VAR) ........................................................................ 26 Appendix C - Residual Terms ......................................................................................................... 28
have left the country with weak public institutions. A year prior to war. the low public wages that make it difficult to attract highly qualified people.5 billion on the back of an increasing productivity. Lebanon’s national accounts are compiled on a yearly basis with actual GDP published with a nine-month to two-year lag. Hence the frequency and scarcity of available data along with lags in releases of macroeconomic indicators constitute the main challenges to estimating GDP. The poor data gathering and mediocre measurement capabilities make it a daunting task for anyone to carry it out. results in several breaks in various series of economic variables. GDP stood at $3. The country has had a dynamic growth in the years leading up to the Civil War with GDP rising 6% per year from 1965 to 1975. Even after the end of the civil war and the starting of the reconstruction phase. Lebanon did not build its statistical capabilities and the country was left without serious national accounts until the early years of the past decade (2000-2009). Consequently. estimating macroeconomic variables was quite avoided in Lebanon. of the importance of having adequate national accounts and up to date information for decision making. 3 . the unavailability of key macroeconomic data prior to 1990s. For example. the official real growth rate of the economy for 2009 is not out yet. a strong Lebanese pound and flourishing banking and tourism sectors. liberal and service-oriented one. and the lack of understanding. Even in the private sector. however. and at best we have guesstimates from the central bank and some international institutions. due mainly to the effects of war on data gathering. on the political front. The national accounts are not being performed at the Central Administration for Statistics. In spite of the importance of the subject.SAL I. The lacks of public investments in institutions. the independence of this unit and its ability to perform its duties without any political interference are questioned. the evolution of the Lebanese economy has followed a cycle of development that shaped Lebanon into an open. The result is that wrong decisions may be taken on both the public and the private sectors fronts for lack of adequate and up to date data. The private sector is not feeling the urgency of investing in such non-for-profit organizations. It is a unit at the prime minister’s office that is responsible of issuing the national accounts.Introduction Macroeconomic statistics have historically been poor in Lebanon. there is no research institution that issues working papers and real scientific surveys on the economic activity. Up until now. Lebanon was not able to put in place a real independent institution responsible for macroeconomic statistics. Going back to before the civil war. In addition.
can bridge this gap. however. For variables other than GDP.SAL In the years that followed the Civil war. is heavily indebted with gross public debt totaling $54. The economy today. represent 70% of GDP. estimating economic growth has a crucial purpose to fill within the large field of policy making and it represents one of the basic problems of statistical analysis upon which authorities rely to set the right policies. a Vector Autoregressive Model with Exogenous variables (VARX). GDP estimates that link near-future growth to current development. but the numbers were often based on incomplete data. Section VI provides a variant of the VAR. led to the restoration of the country’s infrastructure and lifted up the economic activity on all levels. Current GDP levels may provide only insufficient information on future macroeconomic developments. The rest of the paper is organized as follows.4 billion and accounting for 139% of GDP. real estate and banking) sectors that combined. GDP expanded from $1. In this context. and till today. which does not constitute enough data to build the model. present a potential Vector Autoregressive Model (VAR) for GDP estimation that was developed in this paper and address its limitation. For the first challenge we had annual GDP series going back to just 1993. During the conduct of our work. Section VII describe the impulse response of real GDP to shocks on various variables. which offers a better precision in estimating real GDP and overpass the limitations encountered with the VAR. Section II and III provide an overview of the different GDP estimation methods applied worldwide with a focus on the case of Lebanon. we tried to build our model around a set of variables using a vector autoregressive (VAR) approach. In the aftermath of the war. The appendix lay out the complete details of the specifications used in the empirical application. For the second challenge. which does not exist for Lebanon. Section IV and V. it still stands as the principle pillar for economic growth sustainability through services (mainly tourism. the advent of the Hariri Government in October 1992. Lebanese economists were sometimes able to compile a few indicators.3 billion in 1990 to $16. we had monthly series going back to 1993. Lebanon has entered an era where reliable statistics on the state of the economy were usually absent. The private sector was the engine for economic recovery after the war. we try to present a general model to obtain quarterly real GDP growth estimates. 4 . we were faced with two major challenges: the first one was related to available series of data and the second one was to try to come up with an accurate and scientific model to estimate real GDP growth rate.7 billion in 2000. In this paper. Political instability poses as well a severe hurdle for economic growth. Solidere managed the reconstruction of Beirut’s central business district and the stock market reopened in January 1996. So we had to transform the annual GDP numbers into quarterly data to increase the size of our sample and to get real growth rates estimates on a quarterly basis.
g. Andreas Kladroba (2005) carried out a study to compare different methods for a simulated series from an ARIMA (1. Variants of it were proposed by Bournay and Laroque (1979). Lisman & Sandee (1964) and model based methods. Regarding GDP estimation. an example to cite is the Neural Network with an economic application. Stram & Wei (1986). consists of developing a model-based quarterly estimate of GDP derived through bottom-up approach based on actual values of the quarterly components which serve as proxies for selected indicators. and the parametric models. consists of a method that presents a disaggregation scheme (e. The method was adopted by Friedman (1962). The second methodology. quarterly GDP) using information derived only from past and current values of the aggregated series itself (e. called related series (e. annual GDPt-1). and by Chow and Lin (1971).g. The former relies on purely numerical disaggregation technique. with more interest for our purposes. The related series are assumed to be known at the same disaggregation level as the considered disaggregated series (i.g. The many proposed solutions have been reached using one of the following two approaches: The first one being a method which estimates the disaggregated series (e. The latter 5 . This approach is model-based and uses correlated time series to provide estimates of the disaggregated series based on the parameters (𝛽 𝑎𝑛𝑑 𝑢 ) of the aggregated series. Another example of non-model based method.g. consumption of durable goods. The model-based methods use an ARIMA process. annual GDPt. 1) and showed that Chow-Lin is the most accurate for this relative simulated series.e.g. An example would be to divide the annual data into a quarterly figure. Polynomial (1988). The second type. volume of exports). two methods were developed. We distinguish here between non-model based methods. The first one consists of developing indicators relying on a non-model based methodology. is the Polynomial method (1988) which converts the annual series to quarterly figures by fitting a polynomial to each successive set of two points (e. for quarterly GDP disaggregation) based on information which comes from the aggregated series itself and also from other exogenous series. This approach does not involve the use of parameters as no additional exogenous variables are considered.SAL II. quarterly GDP in our case). This approach is known as linear interpolation and it is mostly used for stocks disaggregation. GDP2000 and GDPQ1/2001) to derive a smooth path for the unobserved series. 1. There exists the non-parametric method. adopted in this paper. This approach was adopted by Carriero and Marcellino (2010) and by the Conference Board that constructs the Composite Coincident Indicator (CCI) for the United States as a simple weighted average of selected standardized single indicators.Literature Review Disaggregation methods to obtain quarterly GDP from annual GDP have been extensively considered in econometric and statistical literature. Fernandez (1981) and Litterman (1983).
since a significant portion of electricity is derived from private generators. and broad money.g. according to IIF. imports and exports. growth in tourist arrivals (to substitute for passengers arrivals). The Chow and Lin solution (1971) has been intensively used in National Statistical Institutes. relying on indicators and predictors in different sectors of the economy as proxies to consumption. The reason lies in the practicality of its procedure and in the natural and coherent solution that the model provides to the interpolation problem. Braun (1991) uses a Bayesian vector autoregression (BVAR) to smooth out US production and fill in the missing data for a given quarter. The first estimate for a certain quarter. III. the Error Correction Model (VECM). Other estimates are only indicators for the trend in economic activity. real growth in government consumption (current expenditure minus transfers minus interest payments). and includes five additional indicators: real growth in credit to the private sector (to substitute for growth in deposits). passengers flows. and the Markov Switching Models (MSVAR) which is an updated version of the VAR. provides a graphical trend for GDP over the whole quarter instead of a stock value).SAL include the Vector Autoregressive model (VAR) which is adopted in this paper. Stochastic models were also built and used such as the Dynamic Stochastic General Equilibrium (DSGE) model that provides a continuous path for the estimated variable (e. and trade. The latter follow the BDL’s approach. The model proposed is the first GDP estimate model developed for Lebanon. investment. real growth in government revenues excluding grants. it does not accurately reflect consumption in Lebanon. 6 . The European Central Bank uses log-linear approximations to deal with real-time data set and estimates GDP for the euro area using a VAR model. namely the “Advance” estimate is done by extrapolation based on the monthly trends as incomplete data account for about 30% of the advance GDP estimate. We used the Chow and Lin approach to disaggregate annual GDP into quarterly levels. and real growth in imports of machinery and equipment. The former was developed in 1993 and is called “Coincident Indicator Index”.GDP Disaggregation in the Case of Lebanon We estimate GDP for Lebanon from the demand side. The Banque du Liban (BDL) and the International Institute of Finance (IIF) have designed indices to measure the evolution of the country’s economic activity. cement derivatives (all in volume terms). cleared checks. The Bureau of Economic Analysis (BEA) estimates GDP on an annual and on a quarterly basis. imports of petroleum. It relies on eight indicators: electricity production. Electricity production has been excluded because. We second the argument and therefore exclude it from our model.
we apply the Generalized Least Square (GLS) method to equation (3). equation (1) can be converted to a regression of annual aggregates: 𝑦𝑎 = 𝐶 𝑦𝑞 = 𝐶𝑋𝑞 𝛽 + 𝐶𝑢𝑞 = 𝑋𝑎 𝛽 + 𝑢𝑎 (3) To obtain the Chow and Lin (1971) equation that disaggregates n annual GDP estimates to 4𝑛 quarterly estimates. Let 𝑦𝑞 be a (4𝑛 × 1) vector of quarterly GDP figures to be estimated and let 𝑋 be a (4𝑛 × 𝑘) matrix of 𝑘 exogenous quarterly GDP-related variables. The matrix is defined as: C= 11110000 ⋮ 0 ⋯ 0 ⋱ ⋮ ⋯ 1111 (2) Using subscript 𝑎 to denote annual figures. Chow and Lin have considered the error term 𝑢𝑞 to follow a stationary first order autoregression 𝑢𝑞. Then quarterly 𝑦𝑞 can be predicted using a multiple linear regression of the form: 𝑦𝑞=𝑋𝑞 𝛽 + 𝑢𝑞 (1) Where 𝑢𝑞 is a (4𝑛 × 1) random vector. This yield: ′ ′ 𝛽𝑞 = (𝑋𝑎 𝑉 −1 𝑋𝑎 )−1 (𝑋𝑎 𝑉 −1 𝑦𝑎 ) 𝑎 𝑎 (4) And 𝑢𝑎 = 𝑦𝑎 − 𝑋𝑎 𝛽 (5) 7 (showing no serial correlation) . By assumption.SAL Assume that GDP series are available annually over n years.𝑡 = 𝜌 𝑢𝑡−1 + 𝑒𝑡 (where 𝑒𝑡 is white noise and 𝜎 < 1 ) with having zero mean and covariance matrix 1 𝜌 𝜌2 𝜌4𝑛 −3 𝜌4𝑛 −2 𝜌4𝑛 −1 𝜌 1 𝜌 ⋮ 𝜌3 𝜌 1 𝜌4𝑛 −5 𝜌4𝑛 −4 𝜌4𝑛 −3 ⋯ ⋱ ⋯ 𝜌4𝑛 −3 𝜌4𝑛 −4 𝜌4𝑛 −5 1 𝜌 𝜌2 𝜌4𝑛−2 𝜌4𝑛−3 𝜌4𝑛−4 ⋮ 𝜌 1 𝜌 𝜌4𝑛 −1 𝜌4𝑛 −2 𝜌4𝑛 −3 𝜌2 𝜌 1 V=1−𝜌 2 𝜎 2 𝜌4𝑛 −4 𝜌4𝑛 −3 𝜌4𝑛 −2 Now let C be a (𝑛 × 4𝑛) matrix that converts 4n quarterly observations into 𝑛 annual observations.
The latter variable does also comprise the departures from BIA. These expats are also more resilient than normal tourists to political shocks. we tried through the choice of our variables to have proxies for public and private consumption and investment. “Cement Production”: This variable may be used as a proxy for investment in real estate and infrastructure. The annual residuals 𝑢𝑎 are allocated to the four quarters of the year such that the annual sum of the disaggregated quarterly values 𝑦𝑞 equal to 𝑦𝑎 IV. Non-residents Spending by Credit Cards. These variables will include the Lebanese expatriates who visit Lebanon each year and contribute substantially to its economy.SAL Where 𝑉 = (𝐶𝑉𝐶 ′ ) 𝑎 (6) Using the previous four equalities. Number of Arrivals at the Beirut International Airport and Consumer Price Index. However. we started with the following variables: “Number of Tourists’ Arrival at Beirut International Airport (BIA)”: These tourists have a huge impact on the exports of services and on local consumption. Therefore. We added another predictor for real estate activity that is “lagged construction permits” as the variable’s impact on the economy is slow due to the time lag between the issuance of the permit and the initiation of construction. and the current account position. Petroleum Imports. Hence it gives an idea about both private and public sectors investments in the real estate sector.Choice and Characteristics of Variables Turning back to the model developed in this paper. 8 . the Chow Lin best linear unbiased estimate (BLUE) of quarterly GDP 𝑦𝑞 is derived from: 𝑦𝑞 = 𝑋𝑞 𝛽 + 𝑉𝐶 ′ (𝐶𝑉𝐶 ′ )−1 𝑢𝑎 We obtain an equation of degrees 7 in the unknown ρ with the form: ρ7 + 2ρ6 + 3ρ5 + 4ρ4 + 3ρ3 + 2ρ2 + ρ = ρa 2ρ3 + 4ρ2 + 6ρ + 4 Where ρa is the autocorrelation factor of the annual residual 𝑢𝑎 The estimates of quarterly GDP 𝑦𝑞 in Chow and Lin (1971) model are based on exogenous quarterly variables 𝑋𝑞 and estimated 𝛽 from annual totals for the following five variables: Claims on Private Sector. we seconded this statistic by two other more general ones that are “Total Arrivals” and “International Air Passenger Flows at BIA”.
we adopted two different approaches depending on the nature of the indicators: if the series under construction is that of a flow variable measured over an interval of time. we resort to adding up the three months figures of each quarter to obtain the quarterly value. This could be used mainly as another proxy for consumption. furniture. The figures extracted from the database are available at monthly frequency and so to adjust to quarterly series. “Government Spending”: The considered variable comprises primary spending. daily consumption. The approach adopted in the 9 . The variable also includes loans allocated to businesses for their expansion purposes and thus will in this regard represent investment of the private sector. etc. Our sources of data are mainly the websites of the Ministry of Finance and Banque du Liban (BDL). and Lebanon went through high inflation rates in the 90s. “Cleared Checks”: We thought that the amount or the number of cleared checks will be useful to represent the general economic activity as the increase in the number and amount of cleared checks will probably lead to an improvement in the economic activity. We rebased the CPI to the year 1993 to have a complete time series. in addition to the importance of imports as a proxy to consumption as Lebanon imports more than 80% of its consumption goods. “Petroleum Imports”: This variable will give an idea about the consumption level and more generally about the economic activity as an increase in consumption and economic activity will lead to an increase in the petroleum imports. “Broad Money M3”: This variable represents the impact of monetary policy on the liquidity available in the market. Claims on private sector comprise personal and consumption loans contracted by individuals for their purchases of cars. Primary government spending comprises both current spending that represents public consumption and capital spending that represents public investment. thus public debt service has been excluded. we consider the average value of the three monthly figures of each quarter as the quarterly value. “Imports of Goods (Excluding Petroleum)” and “exports of goods”: These variables serve as proxies to measure the external sector contribution to the economy. If the series considered relates to a stock variable measured at one specific time. in addition to the inflow of capital and remittances. we also included the CPI in our model since high inflation may negatively impact growth. “Consumer Price Index”: Besides using the CPI to deflate our nominal variables.SAL “Claims on Private Sector”: This variable is used as a proxy private consumption and private investment as well. So an increase in M3 will most probably have a positive impact on GDP. “Non-residents spending by credit cards”: This variable is used as a proxy for the additional exports of goods and services that are not being taken into account by the exports of goods that are taken from customs and BDL.
And so X is said to be exogenous.The Vector Autoregressive Model (VAR) The Vector Autoregression (VAR) allows for the forecast of time series and the analysis of dynamic impact of random disturbances on the system of variables.g. The application of the test reduced the number of the endogenous variables of the model to ten out of fourteen initial variables: GDP. Arrivals at the Beirut International Airport. The remaining four variables are exogenous to the model and include: Money Supply (M3). The test shows whether lagged information on a variable Y provides any statistically significant information about a variable X in the presence of lagged X. The procedure consists of decomposing a time series into three components among which the seasonal component and the irregular component. Claims on Private Sector. number of tourists arrival) which are of positive values. We define 𝑌𝑡 as: 𝑌𝑡 = 𝑌1𝑡 . If not. Government Spending. Total Imports (excluding Petroleum). computed by deflating the nominal variables with the CPI. 𝑌2𝑡 . We use the multiplicative seasonal adjustment decomposition model to filter the influence of seasonality. cleared checks. Lag of construction permits. Before proceeding with the model. allows the derived quarterly figure to capture the changes registered over the course of the considered quarter. The seasonally adjusted series are derived using the Census X-12 algorithm used by the US Bureau of the Census. to remove the noise effects that hide the underlying trend in real short term changes. as the magnitude of the seasonal fluctuations vary with the level of the series ( e. We apply the Granger Causality test (1969) to determine the endogenous variables for the model. and Non-resident Spending by Credit Cards. Total Exports. we seasonally adjust the real variables. and Number of Tourists. … . V. Cement Production. 𝑌10𝑡 ′ 𝑊𝑒𝑟𝑒 𝑌𝑖𝑡 . then “Y does not Granger-cause X”.SAL latter case. CPI. The VAR approach considers every endogenous variable as a function of the lagged values of all the endogenous variables in the system. If “Y Granger cause X” and “X granger cause Y” then X and Y are said to be endogenous. 𝑓𝑜𝑟 𝑖 = 1 … 10 𝑎𝑟𝑒 𝑡𝑒 𝑒𝑛𝑑𝑜𝑔𝑒𝑛𝑜𝑢𝑠 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 The VAR (vector autoregressive model) is derived as follow: 10 . Import Petroleum.
24032* 41.796 -1198.408 45. we use the Akaike Information Criterion (Akaike.72106 42. 𝑐2 . 𝜀2𝑡 .92945* 45. we use a general simple model selection approach that yields models with likelihood functions and a finite number of estimated parameters.47313 SC 46.55989* 42. For that purpose.637 -1143.57278 4 5 -1120.058 -1176.3925 43. … .932 * indicates the selected lag order AIC for Akaike information criterion SC for Schwarz information criterion HQ for Hannan-Quinn information criterion The choice of the appropriate length of a lag distribution happens at the lag order that yields the lowest value of the Akaike Criterion. The application of Akaike Criterion yields the following: Table 1: VAR Lag Order Selection Criteria VAR Lag Order Selection Criteria Lag 0 1 2 3 LogL -1303.589 -1109.05549 41.9352 42.648465 44.30909 44. … .SAL 𝑌𝑡 = 𝑐 + 𝛷1 𝑌𝑡−1 + 𝛷2 𝑌𝑡−2 + ⋯ + 𝛷𝑝 𝑌𝑡−𝑝 + 𝜀𝑡 𝑊𝑒𝑟𝑒 𝑐 𝑑𝑒𝑛𝑜𝑡𝑒𝑠 𝑎 (10 𝑥 1) 𝑣𝑒𝑐𝑡𝑜𝑟 𝑜𝑓 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡𝑠 𝑐1 .89904 42.29013 HQ 45.079 41. From the table the proper lag order for the considered VAR reads 4. 𝑐10 𝛷11 (𝑗 ) 𝛷𝑗 = ⋮ 𝛷91 (𝑗 ) ⋯ ⋱ ⋯ 𝛷110 (𝑗 ) for 𝑗 = 1.457 AIC 45. … . 1974) that presents a simple model comparison criterion that uses a penalty term to penalize the log maximum likelihood for lack of closeness.23925 42. 𝜀10𝑡 )′ 𝑤𝑖𝑡 𝐸 𝜀𝑡 = 010×10 𝑎𝑛𝑑 𝐸 𝜀𝑡 𝜀𝜏 ′ = 𝛺 𝑓𝑜𝑟 𝑡 = 𝜏 𝑊𝑒𝑟𝑒 𝛺 𝑖𝑠 𝑎 10 × 10 𝑠𝑦𝑚𝑚𝑒𝑡𝑟𝑖𝑐 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒 𝑑𝑒𝑓𝑖𝑛𝑖𝑡𝑒 𝑚𝑎𝑡𝑟𝑖𝑥 010×10 𝑜𝑡𝑒𝑟𝑤𝑖𝑠𝑒 To determine the lag order of the VAR model.2812 42. 11 .2.3363 42. 𝑝 ⋮ 𝛷910 (𝑗 ) ′ 𝐴𝑛𝑑 𝜀𝑡 = (𝜀1𝑡 .
strengthening the choice of the adopted model selection. For the considered VAR model. So the stationarity of VAR requires the roots of 𝐼10 − 𝛷1 𝑧 − 𝛷2 𝑧 2 − 𝛷3 𝑧 3 − 𝛷4 𝑧 4 to lie outside the unit circle (have modulus greater than one) or equivalently. 𝛺 ) We use the maximum likelihood estimator to estimate the model and derive the parameters.SAL Alternative general criteria exist for model selection. if the eigenvalues of the companion matrix: 𝛷1 𝐼10 F= 0 0 𝛷2 𝛷3 𝛷4 0 0 0 𝐼10 0 0 0 𝐼10 0 Which are those numbers λ that satisfy: 𝐹 − 𝜆𝐼10 = 0 12 . A crucial condition for the VAR model to be valid and consistent requires the covariance to be stationary (i. We test for the stationarity of VAR using the lag operator 𝐿𝑌𝑡 = 𝑌𝑡−1 This translates into: 𝑌𝑡 = 𝐼10 × 𝑌𝑡 = 𝑐 + 𝛷1 𝐿𝑌𝑡 + 𝛷2 𝐿2 𝑌𝑡 + 𝛷3 𝐿3 𝑌𝑡 + 𝛷4 𝐿4 𝑌𝑡 + 𝜀𝑡 𝐼10 − 𝛷1 𝐿 − 𝛷2 𝐿2 − 𝛷3 𝐿3 − 𝛷4 𝐿4 𝑌𝑡 = 𝑐 + 𝜀𝑡 Equivalent to: 𝛷 𝐿 𝑌𝑡 = 𝑐 + 𝜀𝑡 𝑊𝑒𝑟𝑒 𝛷 𝐿 𝑖𝑛𝑑𝑖𝑐𝑎𝑡𝑒𝑠 𝑎 (10 𝑥10) 𝑚𝑎𝑡𝑟𝑖𝑥 𝑜𝑓 𝑝𝑜𝑙𝑦𝑛𝑜𝑚𝑖𝑎𝑙 𝑖𝑛 𝑡𝑒 𝑙𝑎𝑔 𝑜𝑝𝑒𝑟𝑎𝑡𝑜𝑟 𝐿: 𝛷 𝐿 = 𝐼10 − 𝛷1 𝐿 − 𝛷2 𝐿2 − 𝛷3 𝐿3 − 𝛷4 𝐿4 We replace the lag operator L with a scalar z. We proceed by estimating VAR of lag order 4: 𝑌𝑡 = 𝑐 + 𝛷1 𝑌𝑡−1 + 𝛷2 𝑌𝑡−2 + 𝛷3 𝑌𝑡−3 + 𝛷4 𝑌𝑡−4 + 𝜀𝑡 𝑊𝑖𝑡 𝜀𝑡 → 𝑖𝑖𝑑 𝑁(0. The most common are the Hannan-Quinn Criterion (1979) and the Schwartz Criterion (1978).e. This is expected to a certain extent as a lag order of 4 means that the effect of a variable change in a specific quarter will impact GDP for the whole year. time invariant) in order to avoid the formation of explosive roots. the latter two methods yield the same outcome as the Akaike Criterion.
our VAR seems to be stationary (covariance stationary) with its first and second moments time invariants2. Therefore.5 1.5 -1. the result confirms the stationarity of VAR1: Figure 1: Inverse Roots of AR Inverse Roots of AR Characteristic Polynomial 1. 1 The eigenvalues are graphically represented by the inverse roots and their numerical values are computed in Appendix A-2 2 For a detailed approach on how to get the eigenvalues refer to Appendix A-1 3 For further information on VAR refer to Appendix B 13 .0 -0. shows a significant difference between the two values considered for the same period3. the relatively small adjusted R2 (0. called VARX (i.0 0. the backward testing.0 -0.5 0. the eigenvalues are derived in Eviews.5 Therefore. in order to improve the predictability of the model we include the exogenous variables discussed earlier.83) reflects the weak ability of the model to predict a trend. as their inclusion might bring in additional information and contribute for a better trend predictability.0 0.SAL Are less than one.5 -1.5 1.5 0. To achieve that we build a more sophisticated model which is a generalization of VAR.5 -1. In fact.0 -1. However. VAR model holds strongly and passes the stationarity and normality (Jacque-Bera) tests.0 1. Graphically. that provides a comparison between the actual values and the estimated values derived from the model over previous periods. In our model. VAR with exogenous variables).e.
SAL VI. … . we let (𝑋1𝑡 )𝑡 . 𝑋𝑡5 )′ 1 𝛽1 𝜔 = ⋮ 1 𝛽10 ⋯ ⋱ ⋯ 5 𝛽1 ⋮ where (𝛽 𝑖 )𝑖=1. VARX is defined as: 𝑌𝑡 = 𝑐 + 𝛷1 𝑌𝑡−1 + ⋯ + 𝛷4 𝑌𝑡−4 + 𝜔𝑋𝑡 + 𝑢𝑡 Where c denotes a (10 × 1) vector of constants (𝛷𝑗 )𝑗 =1. We proceed by testing for the significance of the exogenous variables(𝑋1𝑡 )𝑡 . (𝑋5𝑡 )𝑡 using the ttest. … .…. VARX is written as: 𝑌𝑡 = 𝑐 + 𝛷1 𝑌𝑡−1 + ⋯ + 𝛷4 𝑌𝑡−4 + 𝜔𝑋𝑡 + 𝑢𝑡 14 .Vector Autoregressive with Exogenous Variables (VARX) We consider 𝑌𝑡 previously defined as: 𝑌𝑡 = 𝑌1𝑡 .𝑝 are (10 × 10) matrices of autoregressive coefficients 𝑋𝑡 = (𝑋𝑡1 .5 are coefficients of exogenous variables in each equation 5 𝛽10 And 𝑢𝑡 is a (10 × 1) vector generalization of white noise with: 𝐸 𝑢𝑡 = 0 ′ And 𝐸 𝑢𝑡 𝑢𝜏 = 𝛺 𝑓𝑜𝑟 𝑡 = 𝜏 with Ω an (10 × 10) symmetric positive definite matrix 0 𝑜𝑡𝑒𝑟𝑤𝑖𝑠𝑒 VARX passes the normality and stationarity tests. We estimate the model using the Multivariate least square estimation method. … . 𝑌10𝑡 ′ 𝑊𝑒𝑟𝑒 𝑌𝑖𝑡 ′ 𝑎𝑟𝑒 𝑡𝑒 𝑒𝑛𝑑𝑜𝑔𝑒𝑛𝑜𝑢𝑠 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 Recalling that the Granger causality test conducted in section IV showed that four out of the fourteen variables under consideration are exogenous. 𝑌2𝑡 . … .…. (𝑋5𝑡 )𝑡 be the time series of the considered 5 exogenous variables. Only two out of the four initial exogenous variables pass the test: money Supply (M3) and lag of construction permits.
92E-12 -2.48E-10 1.026973 4.44E-11 -4.961 0.43E-07 0.41E-11 1.004178 -4.78E-11 6.74E-10 -1.725199 -3.002733 -9.36E-12 -3.02E-15 -1.33E-09 2.03E-06 -2.99E-12 5.99E-09 1.63E-11 3.742172 0.002713 179.973256 5.398408 -0.29E-16 3.90E-10 2.56E-09 5.938883 4. L(TI)= Total Imports Excluding Petroleum.50E-10 3.5 -0.125236 2.99E-12 5.03981 -1.999802 0.58E-11 -1.66E-05 5.38E-06 -1.14E-17 1.80E-05 9.13E+08 0.26E-15 4.81E-06 -0.46E-16 -1.75E-09 4.65E-10 -2.005453 -6371.40663 0.50E-05 6.158157 -0.07E-09 -1.66E-06 2.00177 -0.951557 -0.8713 -0.08E-11 2.02E-08 0.79E-12 2.75E-06 7.966544 1.000104 -0.045937 6.20E-10 -9.846472 -139089.88E-16 -1.61E-08 -0.41E-06 1.00059 2.022697 0.86E-15 9.90E-12 -1.00E-12 -4.36E-07 -0.502 -0.68E-05 -1.71E-11 -1.63E-10 7.67E-07 -0.52E-11 0.146523 -5.34E-09 4.001393 0.98E-11 -5.13E-10 2.42E-06 -1.91E-10 -6.91E-06 1.70012 IP 1.30E-16 9.857 -0.196725 7.734242 3.49E-05 -3.001522 -0.67E-06 -1.22E-06 3.373177 7.25E-07 -1.71E-17 -2.76E-16 1.010927 0.76E-11 -4.53E-05 -3.78E-07 0.69E-18 -3.48E-08 1.61E-09 3.70E-11 2.31E-11 2.996164 0.59E-05 -2.2563 GDP(-1) GDP(-2) GDP(-3) GDP(-4) IP(-1) IP(-2) IP(-3) IP(-4) L(C(-1)) L(C(-2)) L(C(-3)) L(C(-4)) L(CP(-1)) L(CP(-2)) L(CP(-3)) L(CP(-4)) AR(-1) AR(-2) AR(-3) AR(-4) L(GS(-1)) L(GS(-2)) L(GS(-3)) L(GS(-4)) L(TI(-1)) L(TI(-2)) L(TI(-3)) L(TI(-4)) TE(-1) TE(-2) TE(-3) TE(-4) CPI(-1) CPI(-2) CPI(-3) CPI(-4) NRS(-1) NRS(-2) NRS(-3) NRS(-4) M3 LAG(CP.94E-09 1.26E-06 -1.63E-06 -4. S.982 -0.283008 0.255901 -10135.50E-23 0.79E-10 -1.06E-16 1.057933 1.01E-12 3.122233 -3.92E-17 4.25E-11 -1.00E+20 NRS -6.07E-13 7.04E-10 3.49E-10 7.78E-12 -1.63E-11 -7.57E-07 0.88E-08 -1.24E-06 8.94E-07 2.59E-16 8.89E-10 -5.42E-11 6.79E-10 -2.31E-16 1.17E-06 -2.27 0.000557 -0.27E-15 2. L(GS)= log of Government Spending.19E-11 1.09E-13 4.404298 25597.94E-08 -0.04E-20 8.012842 -0.013455 4.14E-11 -2.73E-06 1.004028 1.67E-08 6.96E-10 4.001399 -1.72E-11 9.13E-24 0.50E-08 -5.959992 0.073112 0.21E-17 9.46E-07 -0.000112 6.61E-10 8.28E-11 -4.87E-11 1.59E-16 9.66E-06 6.999239 0.120685 -761.85E-07 1.300915 9792.135737 -6.18E-07 1.004989 -0.291391 0.47E-07 0.02E-10 7.45E-13 0.976854 0.88E-07 1.01E-11 1.62E-10 -7.77E-05 1.99924 0.74E-10 9.49E-05 1.056502 -4.24E-10 1. TE = Total Exports.38E-10 9.46E-18 5.009068 3.74E+22 Where IP= Imports of Petroleum.08E-14 0.93E-07 3.003752 -1.101589 0.77E-17 -6.00E-07 -1.278702 -1.1223 241.84E-19 0.515204 -0.26E-16 7.71E-06 3.35E-18 0.123 0.037485 0.31E-09 -1.73E-09 4.15E-10 -5.14E-10 1.63E-11 -3.42E-09 -2.15E-05 -5.90E-05 -5.01608 -6336.10E-10 -1.51E-05 2.52E-15 0.77E-16 -1.22E-06 2.48E-08 -4.50E-07 0.69 -0.065102 0.056046 -1.89E-12 1.48E-05 0.72E-12 4. AR= Arrivals.154127 -1.163 -0.4756 0.46E-12 3.106243 47.027367 0.26E-07 -8.06E-06 1.43E-09 -6.22E-14 1.071641 0.76E-24 0.008631 -0.71E-16 3.25E-07 1.43E-06 2.4418 0.72E-06 1.399649 0. CPI= Consumer Price NRS= Non-resident Spending by Credit Card.37E+22 CPI 2.88E-16 7.85E-06 2.29E-12 -4.72E-16 7.1511 -0.28E-16 1.000335 2.14E-17 8.18E-11 1.93E-12 5.52E-09 2.30E-07 -1.407277 929.70E-10 5.397 0.20E-05 4.76E-09 -7.34E-08 -1.89E-07 3.72E-07 -3.50E-09 -5.19E-06 -1.40E-12 -4.19E-10 3.83E-05 -1.307033 15171.02034 -0.099503 -5.066573 -1.92E-07 -9.01 0.35E-16 2.57E-11 2.13861 0. R-squared Sum sq.30E-11 9.85E-12 -9. L(CP)= log of Cement Production.956322 0.97E-13 2.91E-12 9.67E-07 -1.39E-09 -3.12E-05 7.000225 8.08E-11 -2.07973 -6971.06E-06 -1.24E-10 -8.71E-13 -1.13E-10 5.15E-10 1.30E-05 3.681804 82001.17E-07 -0.22E-16 -2.313212 0.09504 0.45E-06 -1.02E-07 -4.10E-14 4.57E-06 7.983765 0.5) R-squared Adj.82E+19 L(TI) -1.00E-09 1.95 0.94E-12 4.04E-15 2.48E-07 -3.05E-12 8. L( C)= log of Claims on Private Sector.82E-16 5.E.36E-07 -1. M3= Money Supply 15 .490483 75762.61E-11 2.99E-07 -7.76E-07 0.64E-11 9.23E-12 -1.066242 -7643.002315 -0.564217 1.000323 -0.093154 1115.97E-12 5.86E-09 -5.59E-11 1.24E-15 -8.84E-05 -1.33E-11 -2.46 -0.20E-16 4.001629 0.36E-10 -1.001921 -0.031441 -0.03E-07 -2.25E-06 2.22E-11 -1.36E-09 5.13E-07 3.956322 1.36E+21 L( C) -7.45E-15 2.25E-07 1.25E-16 -2.58E-08 -0.30E-08 0.01E-09 -1.119553 0. CP= Construction Permits.045848 0.02E-17 1.71E-09 -2.63E-16 -3.06E-05 -1.97528 0.228 0.01E-12 -0.17E-07 -3.60E-17 1.076437 1.22E-12 -1.64E-11 -1.24E-07 8.26E-05 6.963344 0.51 -0.SAL Table 2: Vector Autoregressive Model with Exogenous variables (VARX) Vector Autoregressive Estimates L(CP) AR L(GS) -0.12E-08 -0.149105 -9241.578 0.17E-14 6.6 -0.71E-11 1.111198 -9963.71E-05 -3.50E-10 2.04E-17 -1.39E-07 0.08 0.995097 0.60E-12 7.35E-13 -1.82E-05 -2.62E-11 -9.00171 5.296 0.90E-14 4.43E-05 4.66E-16 6.050389 13227.092158 3537.000679 0.250255 -2127.998726 0.57E-08 0.25E-19 0.64E-20 4.76E-10 1.004903 3.205535 -0.39E-15 3.92E-12 4.26E-06 0.000425 482.79E-09 -5.67E-10 -2.258296 0.571442 -0.72E-11 -1.54E-06 2.026443 -1.48E-17 2.49E-10 1.30E-12 -1.008168 -8.55E-07 -1.007104 -6.001155 0.66E-17 2.1 0.000114 -0.973324 6.22E-05 5.59E-10 -4.00E-15 7.003002 489.61E-12 1.99E-16 2.16E-05 -9.002026 0.34E-11 3.48E-05 2.47E-10 -3.94E-18 6.72E-06 0.23E-05 9.41E-12 1.71E-06 4.15E-07 -7.945382 9.97E-16 1.27E-12 8.72E-11 -4. F-statistic GDP -0.000219 0.06E-15 3.50E-05 -7.077129 3910.88E-16 -8.28E+19 TE -2.31E-06 -1.19E-08 1.161266 -1.166738 0.04E-11 -1.008752 417.28E-08 -0.027843 4013.88E-10 5.27E-11 4.022179 0.643978 59460.6875 -0.
122801 0.059532 0.68 accept RESID04 0.864002 0.4649 accept RESID06 0.345286 0.5704 accept RESID02 1.8414 accept RESID10 0.5887 accept RESID03 9. Table 3: Normality test – Jacque-Bera RESID01 JarqueBera Probability Result 1.6492 accept The normal distribution of residuals can be observed graphically: Figure 2: Distribution of residuals 16 .7067 accept RESID09 0.973 accept RESID05 1.581345 0. any change in claims on the private sector during a specific quarter will lead to an increase of GDP over the upcoming year eventhough LC (-1) and LC(-4) have negative coefficients.994863 0.05472 0. The goodness-of-fit test utilizes the information of the sample skewness and sample kurtosis. We proceed by testing for the normality of the error terms of VAR using the Jacque-Bera test.SAL Some of the coefficients’ signs may be unexpected. it shows in the impulse response function that for example.7478 accept RESID07 0. The test is conducted in Eviews and accepts normality at 5%.694423 0. The results of the model as a whole were very satisfactory as at shows. which are the third and fourth moment respectively and are sensitive to small deviations from normality.53185 0.941071 0.6247 accept RESID08 0. However.
The backward testing shows a significant similarity between the series generated by the estimated model and the actual data.005288 0. applied to each residual term of a normal series. VARX is proved to be consistent and significant with normal independent residuals of zero mean.003771 z-Statistic 0. Error 0. 0.003575 10.085.77033 Prob. The test.SAL We have assumed throughout the model residual means of zero. We apply the Durbin Watson statistic test that yields a value of 2. Actual Quarterly GDP figures (in billions of $US) 17 . provides the mean and the standard error of each series and test the significance of each one.89E-05 0. showing that residuals are independent over time. This strengthens the performance of the model and its ability to predict future trends. Here follows the result of the test applied to the residual term of GDP: Table 4: Empirical Distribution Test Parameter MU SIGMA Value 1. The model reveals also very satisfactory when we computed the yearly difference in growth rates between the estimated yearly growth rates and the yearly growth rates: Figure 3: Estimated values in VARX vs. We use the empirical distribution test to prove it.040615 Std. The following chart shows the estimated quarterly GDP figures plugged against the actual quarterly GDP-Chow Lin figures.9971 0 The application of the test to the remaining nine residual terms reveals that the all ten variables are all normally distributed with mean zero.5%. We proceed by testing for the serial correlation of residuals. The model holds strongly with a high adjusted R2 of 97.
SAL Table 5: Estimated values in VARX vs. keeping other variables fixed through the Choleski decomposition transformation: 18 .3% 3. and Imports of Petroleum.3% 5.6% -0.2% 7.5% 7. We measure the impulse response of three endogenous variables on GDP: Claims on Private Sector. The following graph tracks the response of real GDP over time to a positive shock of one standard deviation on Claims on Private Sector.5% 1.5% 7.6% 3.1% 9. we add up the estimated values of GDP growth for the four quarters of the considered year.4% 3.0% 0.6% 7.5% 1.2% 3.2% 4.2% -0.1% -2.5% 9.2% 1.6% GDP actual yearly growth 6. Passenger Flows.5% 5.3% 8.3% 8.6%.1% -2. The shock on claims takes place in Quarter 1.0% 3.0% 0. For 2009. VII.6% 1.3% 4.0% 3. real GDP growth is estimated at 8.5% Difference in Basis Points 23 3 -29 39 -21 17 -59 14 -24 -9 -2 5 40 -1 -6 To obtain the estimated yearly growth rate of GDP for a given year.Impulse response We generate impulse response functions for the VARX model to assess the sensitivity of estimation performance and certain policy analysis. Actual Annual GDP figures GDP estimated yearly growth 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 6.
will be targeted towards either households or investors. The economic interpretation of the positive relationship between Claims on Private Sector and GDP is that an increase in the amount of loans extended by banks to the public increasing by that the claims on private sector . A flattening growth of GDP is observed in the second quarter (graphically. from 2 to 3). In the former case. Thus. consumers will use the loans to spend more on nondurable and durable goods. In the former case. businesses will make use of the loans to start up or expand their businesses. over four quarters. Another impulse response function worth observing is that of GDP to a one standard deviation positive change in Number of Arrivals at the Beirut International Airport. a rise in private claims will impact positively GDP. regardless of the targeted audience. will impact real GDP positively over the course of a year (graphically. A plausible explanation is seasonality. from 1 to 5).SAL Figure 4: Accumulated Impulse Response of Real GDP to a one Positive Standard Deviation Shock on Claims on Private Sector An external positive shock that would increase Claims on Private Sector. The Choleski decomposition yields the following: 19 .
SAL Figure 5: Accumulated Impulse Response of Real GDP to a Positive Standard Deviation Shock on Arrivals A shock that would lead to a higher Number of Arrivals would positively impact real GDP over the next three quarters (graphically. We have also tracked the impulse response of GDP to a positive shock of a one standard deviation on Imports of Petroleum. The reason lies in the fact that more tourists would boost the level of spending lifting by that the level of economic growth. from 1 to 5). The outcome entails the following: Figure 6: Accumulated Impulse Response of Real GDP to a Positive Standard Deviation Shock on Imports of Petroleum 20 .
That positive relationship between Imports of Petroleum and real GDP can be mostly reflected through the boost in consumption level. the margin of error in the estimation process would narrow. +40bps during the sample period extended from 1995 to 2008. and more precisely GDP growth. If the Lebanese government intensifies its effort and develop further its National Accounts to have economic data. would cause an increase in real GDP over the course of a year (graphically. the gap between the actual and fitted annual values lies in the range of -59bps. The upward trend in GDP will start to decelerate over the fourth quarter (graphically.Conclusion The VARX model developed in this paper indicates that when looking at historic GDP. published on a quarterly basis then we can assess better the economy and provide an even more accurate model. holding other variables constant. The impulse response function conducted in the last section strengthens the estimation capabilities of the model as the results proposed capture the economic significance and are in line with the outcome of VARX. from 4 to 5) due to the fading of the effect of the shock. This suggests a high level of accuracy and precision and validates the model as a useful tool in guiding estimation judgments’.SAL A rise in Imports of Petroleum. The precision of the estimated values would increase significantly in the latter context as the disaggregation stage would be omitted and thus. and partly through the rise in enterprises and manufacturing activities. from 1 to 5). A more interesting and effective insight could be gained by making actual quarterly GDP growth available. VIII. 21 .
B. B. Institute for Statistics and Econometrics. (2008). (1987). (1994). 2.. Caro. The Review of Economics and Statistics. Monsell. 63. Guerrero. “Taking the Pulse of the Economy: Measuring GDP”. R. and Extrapolation of Time Series by Related Series”. 359-379. (2005). “A Metohdological Note on the Estimation of Time Series”. European Commission. Chow.. D. M. A. M. B. 22. Feijoo.55. 4. R. 22 . Hedhili. C. 4. Bera. 53. and Kapetanius. 1. R. 372-375. G. J. European University Institute. A. 198-216. Computational Statistics and Data Analysis. C. “Methods for Quarterly Disaggregation Without Indicators. and Seskin. and Quintana. H. S. Chen. M. Marcellino.C. E. “A recursive ARIMA-based Procedure for Disaggregating a Time Series Variable using Concurrent Data”. “New Capabilities and Methods of the Census X-12 Seasonal Adjustment Programs”. and Jarque. 16. “Lag Selection in Subset VAR Models with an Application to a US Monetary System”. 127176. K. M. “Time Series Analysis” Princeton Universtiy Press.P. D.. and Otto. A. A.S. R.. and Lutkepohl. (2003). 2. “An Emperical Comparison of Methods for Temporal Disaggregation at the National Account”. “Econometric Analysis with Vector Autoregressive Models”. W. Journal of Economic Perspectives. Landefeld.SAL References Bell. (2005). 471-478. and Trabelsi. Journal of Business and Economic Statistics. G and Lin. “A Test for Normality of Observations and Regression Residuals” . (2000). Bureau of Economic Analysis. 63-78. and Martinez. Hamilton. (2010) “Forecasting large datasets with Reduced Rank Multivariate Models”. (2007).. I. International Statistical Review. Fraumeni. Luetkepohl. L. C. Test. (1998). Luetkepohl. 2. Bruggemann. Springer-Verlag Press. Carriero. F. H. V. R. A. “Best Linear Unbiased Interpolation. Fernandez. (1971). Distribution. 43.. J. 2. M. D. “A Polynomial Moethod for Temporal Disaggregation of Multivariate Time Series”. B. D. 163-172. H. “New Introduction to Multiple Time Series Analysis”. European University Institute. Findley. The Review of Economics and Statistics. (1995). (1981). a Comparative Study Using Simulation”. (2007). Chen.
“Lag Length Selection in Vector Autoregressive Models: Symmetric and Assymetric Lags”. S and Salazar. Dsicussino Paper 127. J. Proeitti. (1999). Ozcicek. The Econometrics Journal. (2006). (2004). “An Indicator of Monthly GDP and an Early Estimate of Qaurterly GDP Growth”.SAL Mitchell. M. 4. 357-372. “Temporal Disaggregation by State Space Methods: Dynamics Regression Metohds Revised”. Applied Economics. R. 3. J.. O. Weale.. T. National Institute of Economic and Social Research. 9. Smith. Wright.. 23 . L. R. 517-524. 31.
However.VAR Stationarity 1.SAL Appendix A . ⋯ 0 ⋱ ⋮ ⋯ 0 𝛺 𝑏𝑒𝑖𝑛𝑔 𝑡𝑒 𝑣𝑎𝑟𝑖𝑎𝑛𝑐𝑒 𝑐𝑜𝑣𝑎𝑟𝑖𝑎𝑛𝑐𝑒 𝑚𝑎𝑡𝑟𝑖𝑥 𝑜𝑓 𝜀𝑡 𝐴𝑛𝑑 𝑄(40×40) = The effect of 𝑌𝑡+𝑗 of a one unit increase in 𝜀𝑡 indicates whether VAR(4) is explosive or stationary: If the dynamic multiplier is between 0 and 1.Conditions for stationarity The value of the scalar system of VAR (4) at time t is given by the following dynamic equation: 𝑌𝑡 = 𝑐 + 𝛷1 𝑌𝑡−1 + 𝛷2 𝑌𝑡−2 + 𝛷3 𝑌𝑡−3 + 𝛷4 𝑌𝑡−4 + 𝜀𝑡 For our purpose it is more convenient to rewrite 𝑌𝑡 as a first order difference equation in a vector 𝜉𝑡 . then the absolute value of the effect decays geometrically towards zero and stationarity is satisfied. 24 . consider the following first order difference equation which defines VAR (4): 𝜉𝑡 = 𝐹𝜉𝑡−1 + 𝑉𝑡 𝑊𝑒𝑟𝑒 𝐸 𝑉𝑡 𝑉 ′ = 𝜏 𝛺 ⋮ 0 𝑄 𝑓𝑜𝑟 𝑡 = 𝜏. 040×40 𝑜𝑡𝑒𝑟𝑤𝑖𝑠𝑒. if the dynamic multiplier is lower or higher than 1. Define: Vector 𝜉𝑡 by 𝜉𝑡 = 𝑌𝑡 𝑌𝑡−1 𝑌𝑡−2 𝑌𝑡−3 Matrix 𝐹 for a 4th order by 𝐹 = 𝛷1 𝛷2 𝐼10 0 0 𝐼10 0 0 𝛷3 𝛷4 0 0 0 0 𝐼10 0 𝐴𝑛𝑑 𝜀 𝑡 0 0 0 Vector 𝑉𝑡 by 𝑉𝑡 = Under the above notation. then the system would either exhibit explosive oscillation or the dynamic multiplier would increase explosively over time.
001972i -0.31e-06 .366042i 0.054897 .002089 0.366042i -0.007141i 0.426946 0.081447 .70483 0.001802 0.332832 + 0.0.6452 0. the VAR is stationary if the eigenvalues of the matrix 𝐹 lie inside the unit circle.001805i -0.0.883836 0.430106 .000838 .007035 .332832 .0.0.0.001809 -3.00279 0.31e-06 + 0.00209 0.001915 .924374i 0.000890i 0.426946 0.001972i 0.001805 0.927955 0.0.000837i -0.0.0.000838 + 0.0.000890i Modulus 0.001915i -0.564782 0.001973 .054897 + 0.847262 0.847262 -0.430106 + 0.60437 0.007141i -0.000837i 0.70483 0.0.0.001973i 0.000837i 0.001915i 0.115952 + 0.00279 0.Numerical values of the inverse roots Root 0.000890i -0.002089 0.462054i -0.001258 0.423402i -0.007035 .010024 0.0.00279 0.0.001259 0.002089 0.0.081447 + 0.000890 + 0.009877 Root -0.423402i -0.001973 + 0.006933i Modulus 0.000890 + 0.184468i -0.001258 No roots lies outside the unit circle VAR satisfies the stability condition 25 .000837 .857306 + 0.001809 0.532251 + 0.001915i -0.001805 0.001259 0.634695i -0.001973 + 0.380533 0.0.00209 0.001973i -0.0.007035 + 0.001914 + 0.564782 0.000837i 0.214921i -0.532251 .000890 .6452 0.634695i 0.927955 0.115952 .001914 .857306 .883836 0.0.462054i -0.SAL The analytical characterization of 𝜕𝑦 𝑡+𝑗 𝜕𝜀 𝑡 is obtained in terms of the eigenvalues of the matrix𝐹 which are those numbers 𝜆 for which: 𝐹 − 𝜆𝐼10 = 0 The eigenvalues for VAR (4) are the solutions to: 𝐼10 𝜆4 − 𝛷1 𝜆3 − 𝛷2 𝜆2 − 𝛷3 𝜆 − 𝛷4 = 0 Consequently.00209 0.001805i -3.60437 0.006933i 0.002089 0.924374i 0.001915i 0.00279 0.000837 + 0.001973 .010024 0. 2.0.009877 0.000890i 0.184468i -0.214921i 0.000890 .380533 0.001915 + 0.007035 + 0.001802 -0.00209 0.
39 -51086.031739 -4.31E-02 -7.78E-08 0.012843 0.674433 0.00216 -0.3 -360848.387771 0.E.193558 IP -8.69 44286.000447 0.147671 2.51E+04 -41005.010638 0.11E-06 2.068221 -0.443 13133.57E-09 1.000451 -1.47375 -1.918256 -4.592 -536.006653 0.54 -49878.13E-06 6.1 1121131 -1226677 -760287.1 -1139715 67963.85E-02 4.78E+00 -5.020622 0.26752 0.60E-08 -1.06E-08 2.00E-01 -2.67E+06 3.378437 -0.49E-08 8.22E-07 -1.63E-02 -1.87 -117448.093 3365.197097 -1.85918 -0.00154 -0.47E+00 -1.7 -36655.62E+04 1.00192 -1.04E-07 -7.009548 0.62E+04 106340.386387 -0.22E-01 0.49E-07 1.22E+00 3.519215 5.074169 0.998856 0.00E+10 38747.117141 TE 1.19455 -0.15 33893.33E-08 -2.040944 -0.60E+11 134131.19E-06 -1.05E-07 -1.001569 -0.45E+00 8.033348 0.006904 -0. TE = Total Exports.639735 2.37E+01 -5.159022 1.806437 -1.23E+05 -187094.157105 -0.516981 0.22E-06 -1.79E-01 4.894091 -3.883946 -3795.07E-07 1.9 100592.126369 -0.31E-08 5.63E-09 5.15E-03 -0.6 -8.085286 0.5 -0.34E-07 5.868047 8.87365 0.9 221583.204981 -0.15062 -2.48E-06 7.58E-08 -3.45E-07 2.85E-05 -5.181593 0.1 -23706.49E+00 -5.083682 -115304 552813.00103 -1.036373 -3.1214 0.46267 -5.38E-07 -5.91E-06 -0.2 908304.083156 0.85E-07 0.115509 -0.29E-07 -1.11327 0.57E+04 116047.35E-01 -0.09E-06 5.127665 -0.484 20874.001856 -0.109077 -0.06 181250. L(CP)= log of Cement Production.22E-06 1.83E-07 -7.079967 -1.067 2504.81E-02 -2.214389 0.200527 0.77 108320 -74735.19E-07 -3. CPI= Consumer Price NRS= Non-resdient Sepnding by Credit Card 26 .24 -42295.972706 7.0913 -0.92E-07 9.72E-08 3.36 46300.51584 L(GS) -0.639 0.134976 0.017177 0. F-statistic 0.2 -8.53 -75452.91E-08 -2.528391 1.152156 0.43E-08 -5.57E+00 -3.389103 2468.68 -261397 132021.65117 0.01E-07 1.326528 -0.897052 0.004934 0.216284 0.21E+10 64078.027533 -0.71E-09 4.173079 -0.530636 0.411937 -1.86E-07 3.91308 Where IP= Imports of Petroleum.766 -5178.36E-01 -2.52E+06 1.347266 0.38E+06 1.452003 0.34E-08 1.052666 -0.7916 0.328628 -0.04238 -0.72E-09 1.97E-01 -0.003047 0.736882 13.941098 0. L( C)= log of Claims on Private Sector.138027 -0.122825 2.696302 3.41484 -2.99E-08 -4.05E-07 0.002275 -1.13282 -1.349809 -19.13E+05 -1408977 95779.39E-07 -4.26E-10 -5.28E-08 -1.26E+05 4.31E+06 9.99005 0.07E-07 -1.095271 -5.108448 -0.06 -65158.77E+04 -82306.18E-08 0.55E-07 2.29E-06 2.004419 0.52 -380836.66E-02 0.95E-07 3.48E-06 -9.004734 0.746405 AR 33974.000787 -2.970647 185.84915 3.10E-06 -2.59E-09 3.65E-08 -1.89E-06 1.63 -0.12956 1.509621 -0.30E-06 -3.40E-05 0.12E+00 5.02804 0.05E-07 0.21E-01 -7.658736 L(TI) -6.825251 1.56E-06 9.035015 -0.47E-08 0.74305 -2.43E-06 -7.157053 26008.56E-08 -1.51E-07 0.505761 0.215236 0. L(GS)= log of Government Spending.61E-07 -0.127589 0.812779 0.730947 -0.432651 -0.922651 0.088467 -8.76E-06 1.058805 -1.83E-08 1.070768 0.24E-08 1.00E+05 -4.344 L(CP) 0.04167 51.468519 L( C) -8.40E+05 6.764086 2.003125 -6.63 4534.02652 NRS 1.006276 -0.09E-02 -5.66 0.98E-03 2.73E-06 8.44 0.32E-05 -1.085739 0.021964 -0.83E-02 -0.0069 -0.00124 -0.00E-07 4.8 0.072014 0.771821 0.95527 0.634218 0.3 54.68176 1.70E-02 -2.68E-05 2.45E+06 -48072.16 -49614.75E+00 6.865113 5.84E-01 -1.71E+12 620885.7 -109351.009678 1322.72E-02 -3.233843 0.SAL Appendix B .034349 -0.303347 0.746355 2.35 -41335.54E-07 -1.13177 6.57E-06 1.68 11776.231274 -1.383321 1. R-squared Sum sq.26 9.7 -4.763847 0.19E-05 3.560868 0.030197 1.56913 -742.929356 0.647581 -5.99 2226.013 1054.948865 0.5 2250200 -2133070 -6.002028 -0.320827 0.061382 0.Vector Autoregressive Model (VAR) 1.7748 -679.153024 0.95199 CPI 1.59E-01 -1.507 19830.43E+00 -2.08248 -7.21158 -9.53E+04 -2.80E-08 2.2 60940.18E+00 0.006084 -0.30E-01 1.318388 0.24E-02 3.94E+06 -2993809 -30840.06189 0.21 -280842.58 -18344.008537 -0.082151 6.016474 0.829638 0.854846 -3.83E-06 -6.77E-08 0.089917 0.76E-08 0.47E-07 -1.67E+06 1.48E-08 6.9 0.122729 0.014374 0.58 1.06E+00 6.06 -0.13E-07 8.015551 0.48E-08 -2.00229 -0.5526 -0.42 5154.883129 0.026194 0.2025 0.011818 0.17E-06 0.1 4.0351 3.20E-06 -1.389259 0.88E-05 4.88E-07 1.213881 -6.80E-05 2.080328 -0.198073 -0.06E-02 4.67E+00 4.764728 5.990748 0.426073 5.17E-02 -4.5 -210598.53E-06 -5.22E-01 -2.035629 -0.113608 0.168992 0.999612 0. AR= Arrivals.038688 0.27E-07 2.82E-08 4.032774 0.9 -164220 183585.40E+05 -41034.83 4.765241 -0.589063 0.05E+05 6.74E-09 8.82624 1.001939 3.97E-02 -2.96E-07 0.68E-01 -1.24E-07 0.29E-06 -1.013789 -0.0662 -0.012667 0.001873 0.54 -54545.035888 0.494241 2.79E-01 -2.45E-01 3.156 -170.227829 5.49E+01 4.15E+00 7. S.17E-08 -9. L(TI)= Total Imports Excluding Petroleum.48 -0.76E-07 1.443346 23483.27 71259.09E-08 5.35 10.The model Vector Autoregressive Estimates GDP GDP(-1) GDP(-2) GDP(-3) GDP(-4) IP(-1) IP(-2) IP(-3) IP(-4) L(C(-1)) L(C(-2)) L(C(-3)) L(C(-4)) L(CP(-1)) L(CP(-2)) L(CP(-3)) L(CP(-4)) AR(-1) AR(-2) AR(-3) AR(-4) L(GS(-1)) L(GS(-2)) L(GS(-3)) L(GS(-4)) L(TI(-1)) L(TI(-2)) L(TI(-3)) L(TI(-4)) TE(-1) TE(-2) TE(-3) TE(-4) CPI(-1) CPI(-2) CPI(-3) CPI(-4) NRS(-1) NRS(-2) NRS(-3) NRS(-4) R-squared Adj.285851 8.01E+06 -10091563 -1128596 -947683.03E+00 2.10449 -0.586348 0.
29% 3.48% 0.95% 3.37% 3.05% 5.Estimated values in VAR vs.49% 9.45% 1.75% 3.81% 5.Estimated values in VAR vs.00% 6.88% 4. Actual Quarterly GDP figures (in billions of $US) 3.54% 5.24% 7.52% 4.59% 6.87% 3.14% 4.74% 1.24% -0.53% 3.59% -0.41% 9.34% 3.00% 0.30% Difference in basis points 405 489 -214 -85 -233 -217 -57 -38 -65 274 -81 -482 -209 325 27 . Actual Annual GDP figures GDP estimated yearly growth 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2.SAL 2.78% GDP actual yearly growth 8.48% 1.45% 1.14% -2.59% 7.
000 -.0000000 -.0E-10 1994 1996 1998 2000 2002 2004 2006 2008 -.0E-11 4.04 -.00002 -8.04 .10 1994 1996 1998 2000 2002 2004 2006 2008 .0E-11 -.Residual Terms 1.0E-11 1994 1996 1998 2000 2002 2004 2006 2008 .0000005 -5.12 1994 1996 1998 2000 2002 2004 2006 2008 -.05 -.005 0 .00003 1994 R_PURCHASED_AMOUNT Residuals 1996 1998 2000 2002 2004 2006 2008 28 .000000 2.015 .12 .Plot REAL_GDP_CHOWLIN Residuals .0000010 -1.000 1994 1996 1998 2000 2002 2004 2006 2008 -.0E-10 .00003 .000 .0E-11 -.0E-10 -.005 -2.000 .0E-10 .0000005 0.00001 -4.000 -.0000015 .0E+00 .00001 0.00 -2.000008 1994 1996 1998 2000 2002 2004 2006 2008 -4.0000010 5.0E-10 LOG(R_IMPORTS) Residuals .15 .010 2.05 0.0E-10 -.0E-11 .08 -.010 -4.0E-11 LOG(R_CLAIMS) Residuals .015 1994 1996 LOG(R_SPENDING) Residuals 1.10 LOG(CEMENT_PRODUCTION) Residuals ARRIVALS Residuals 4.0000015 1994 1996 1998 2000 2002 2004 2006 2008 1994 1996 R_EXPORTS Residuals 1998 2000 2002 2004 2006 2008 1998 2000 2002 2004 2006 2008 CPI Residuals 8.08 .00000 -.000004 .0E-11 .0E+00 .000008 IMPORT_PETROLEUM Residuals 6.000004 -.00002 4.00 -.0E-11 -6.SAL Appendix C .0E+00 .
SAL 29 .
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue listening from where you left off, or restart the preview.