FORECASTING THE ROMANIAN GDP

Transcription

1 6. FORECASTING THE ROMANIAN GDP IN THE LONG RUN USING A MONETARY DSGE 1 Pere CARAIANI Absrac In his sudy, I esimae a moneary DSGE model using Bayesian echniques and I use he esimaed model o forecas he Romanian GDP in he long run. For he period, he forecass wih he model confirm he presen consensus among he economiss abou a growh poenial of 5 o 6% for Romania. In he long run, he model forecass a sable annual growh rae of abou 4.9%. Keywords: forecasing mehods, DSGE models, Bayesian mehods, real business cycles JEL Classificaion: E6, C68 1. Inroducion One of he fundamenal issues in macroeconomics is as accurae a forecasing as possible of he fuure dynamics of he economy and of he main macroeconomic variables. For an emerging economy like Romania his is even more imporan, given he need of adoping economic policies ha encourage he economic developmen. There are several approaches in forecasing he dynamics of Romania using macroeconomic models. One of he more recen sudies which approach he economic growh in Romania is ha of Croioru and Târhoacă (23). They calibraed a CGE model for Romania which hey used in order o esimae he growh rae in he medium run. They obained a growing growh rae, which ends oward 6% a he end of he forecasing period, namely 21. We can also noice Albu and Roudoi (23) sudy regarding he applicaion of he producion funcions o he analysis of he economic growh in Romania. 1 This research is par of he CEEX projec: Economic Growh, Employmen and Compeiiveness in he Knowledge-Based Economy, 25-28, coordinaed by he Insiue for Economic Forecasing, Romanian Academy. This paper was finished in Sepember 28, before he effecs of he global financial crisis sared o be fel in Romanian economy. Researcher a he Insiue for Economic Forecasing, Romanian Academy; Caraiani@ipe.ro. Romanian Journal of Economic Forecasing 3/29 75

2 Insiue of Economic Forecasing Caraiani (27a) simulaed a Solow model in order o do long run forecass for he GDP growh rae. He calibraed he Solow model using he daa regarding he dynamics of he Romanian economy during he ransiion period. He showed ha for he 22 horizon he growh poenial of he Romanian economy was of almos 4%. We can also noice he yearly forecass wih he Dobrescu macromodel. A descripion of he Dobrescu model can be found in Dobrescu (26). In he field of macroeconomics, he dominan paradigm oday is ha of he dynamic sochasic general equilibrium models (DSGE, hereafer). The DSGE approach is based on micro foundaions, raional expecaions, opimizing agens (household, firms), and differen ypes of rigidiies, eiher nominal or real. The DSGE approach is a coninuaion of he modeling approach sared by Kydland and Presco (1982), who proposed he real business cycle models (RBC, hereafer) as an alernaive o he radiional Keynesian models. One of he firs sudies o apply a DSGE model o he forecas of he Romanian economy is ha in Caraiani (28). He esimaed a RBC model for he Romanian economy using Bayesian echniques, and used he esimaed model o forecas GDP for he period. He esimaed an annual growh rae of abou 6% for his period. In he presen sudy, I propose he use of a DSGE model wih cash in advance consrain in order o forecas he Romanian GDP. The moneary DSGE models are models which exend he sandard RBC by inroducing money. The inroducion of money in a DSGE model is a opic which coninues o be debaed, as here is no agreemen on he righ inroducion of money in a DSGE model. One possible approach o his problem is he inroducion of money in he uiliy funcion (MIU, hereafer). This approach assumes ha some uiliy is derived from he presence of money and his jusifies he incorporaion of moneary balance in he uiliy funcion. This model is characerized by super-neuraliy ha is, no even ha he model has he neuraliy propery, bu, as Walsh (23) noiced, even he changes in he money growh rae do no have real effecs. However, he MIU approach has one severe limiaion, as Walsh (23) showed, namely ha i does no acually answer o he quesion of why money produce uiliy. One model ha depars from he MIU approach bu answers beer o he moneary issues is he CIA (cash-in-advance) model. The CIA models require ha he moneary balance are possessed in order o finance cerain ypes of goods, ha is, money are required o buy cerain goods, Walsh (23). The moneary CIA models were used in several sudies in order o invesigae he relaionship beween money and producion, or he relaionship beween inflaion and producion, or he cos of inflaion. Thus, Cooley and Hansen (1989) incorporaed money in a real business cycles model using a cash-in-advance consrain. They used he model in order o sudy he impac of inflaion on he welfare, and also o sudy he long-run differences beween low inflaion economies and high inflaion economies. They showed ha 1% inflaion implies a welfare loss of.387% of GDP. Cogley and Nason (1994) used several moneary models, including a CIA model, in order o sudy he degree o which hey could replicae he feaures of he American economy. Alhough hese models are characerized by weak propagaion 76 Romanian Journal of Economic Forecasing 3/29

3 Forecasing he Romanian GDP mechanisms, as he auhors noiced, hey can relaively easy simulae he dynamics of he American economy. On a opic closer o he presen sudy, Chari, Jones and Manuelli (1995) sudied he relaionship beween inflaion and producion using several mehods, including also a CIA model. They showed ha he resul according o which here was a negaive correlaion beween inflaion and economic growh i was no confirmed when aken ino consideraion he role of he banks in he financial sysem. In a sudy which proposed a new economeric mehod o evaluae he DSGE models, Schorfheide (22) evaluaed wo moneary models, namely he CIA and he PAC (porfolio adjusmen cos model) using a loss funcion. His approach was based on he Bayesian mehodology. He answered o he classical quesions regarding he predicion of he moneary models, namely of how well hey reproduce he sylized facs regarding inflaion and producion, of how well he model reveals he correlaion beween inflaion and producion, and of how well such models could be used o simulae he impac of shocks in nominal growh rae on he economic growh. He concluded ha he CIA model is superior in poserior odds raios and in he qualiy of he esimaion. Moreover, hese models confirmed he negaive relaion beween producion and inflaion, bu he correlaion appeared as overesimaed. The disadvanage of he CIA model was ha i could no reproduce he posiive effec of he modificaion in money supply on he producion. This paper sars from he resuls in Schorfheide (2) and proposes iself o apply a moneary CIA model for he Romanian economy. The model is presened in he second secion. The hird secion presens he daa used for Romania and discusses he resuls of he Bayesian esimaion. In he fourh secion I use he model o forecas he quarerly GDP. The las secion concludes and oulines some possible exensions of his sudy. 2. The Model I presen in his secion he CIA model as oulined in Schorfheide (2), and also used in Cogley and Nason (1994). The model is composed from a household secor, a firms secor, a banking secor and a moneary auhoriy. The problem of households is o choose consumpion C, he work ime H (which is normalized beween and 1), and he money supply M +1, as well as he deposis D so ha i maximizes he expeced oal lifeime uiliy given by: max E β [(1 φ)ln( C ) + φ ln(1 H )] (1) = The consrains are given by he following equaions: P C M D + W H (2) which signifies he CIA consrain D (3) Tha shows ha he household do no borrow from banks. Romanian Journal of Economic Forecasing 3/29 77

4 ( M D + W H PC ) + RH D + F B Insiue of Economic Forecasing M + 1 =, + (4) which is he budgeary consrain. The firms maximize he presen expeced value of fuure dividends. The maximizaion problem consis in he opimal choice of dividends F, capial sock K +1, labor force demand N and loans L from banks in order o maximize: + 1 F max E β (5) = C+ 1P + 1 Under he consrain given by: α 1 α F L + P [ K ( A N ) K+ 1 + ( 1 δ ) K ] W N L RF, (6) Here β is he discoun facor, δ he depreciaion rae, while α is he capial share. And by: W N L (7) The hird ype of agen, he represenaive financial inermediary, solves he following problem by opimally choosing B, L and D : + 1 B max E β (8) = 1 C+ 1P + 1 The banks in he banking secor receive cash dividends from he households, and money from he cenral bank. The money are used o give credis, L, o firms, and he reurn of he credis is R F, -1. The behavior of banks is characerized by he following wo equaions: L X + D (9) which sands for he equilibrium condiion in he credi marke. The following equaion expresses he bonds: B = D + RF, L RH, D L + X (1) The model is closed by adding he equilibrium condiions for he labor force: H = N (11) for he money marke: P C = M + X (12) and for he goods marke: ( ) α 1 K 1 δ K K ( A N ) α C ( ) = (13) In he equilibrium he following relaion sands: R H, = RF, (14) The producion funcion is a Cobb Douglas one and is given below: α 1 ( A N ) α Y = K (15) 78 Romanian Journal of Economic Forecasing 3/29

5 Forecasing he Romanian GDP Two ypes of perurbaions are inroduced, one sanding for he echnological shocks, which follow: ln A = γ + ln A 1 + ε a, (16) While he second sands for he moneary shocks: ln m = (1 ρ ) ln m * + ρ ln m 1 + ε m, (17) Here ρ is he auocorrelaion of he innovaions in he money growh rae while m * is he uncondiional mean of he money growh rae. γ is he deerminisic rend of he echnology growh. 3. Daa and he Esimaion of he Model For he esimaion of he model, I used wo ime series, namely GDP and inflaion. Boh series are used a quarerly levels. For GDP, I used he quarerly GDP in 1995 consan prices. The inflaion daa is given by he quarerly GDP deflaor. Alhough boh series feaure a rend, I used he ime series as given (I only de-seasonalized hem), as he model I use allows for he presence of a rend in he producion and inflaion. * The se of parameers o be esimaed is given by{ α, β, δ, γ, m, ρ, φ, σ a, σ m}. Several of hese parameers were calibraed using he resuls in Caraiani (27b). The elasiciy of producion wih respec o capial α, he capial share, was calibraed o.4. The depreciaion rae δ was fixed o.24. Using he resuls in Caraiani (27b), I also could se he discoun facor β o.98. The esimaion was done using Bayesian echniques. I used wo Meropolis Hasings chains, each one of 5. exracions. The convergence crieria as proposed by Brooks and Gelman (1998) showed ha he convergence was achieved. In Appendix, he resuls of he esimaion are presened. Table 1 Bayesian Esimaion Resuls Prior Poserior Confidence Confidence Prior Sandard Parameers mean mean inerval inerval disribuion deviaion γ Normal.3 m * Normal.7 ρ Bea.5 φ Bea.5 σ_a Gamma Inv. Infinie σ_m Gamma Inv. Infinie Source: Auhor s own compuaions. Table 1 presens he resuls of he esimaion. From Appendix, we also can noice ha he poserior disribuions are characerized no by variabiliy, bu also by significan differences wih respec o prior disribuions. This shows ha he esimaion resuls were significanly influenced by he disribuions of he daa used. Romanian Journal of Economic Forecasing 3/29 79

6 Insiue of Economic Forecasing For he ρ coefficien, which characerizes he auocorrelaion of he money growh rae, we go a quie high value, implying a high persisence in he money injecion shocks. The γ parameer, which refers o he deerminisic componen of he echnology process, was esimaed a a higher value han he prior mean, underlining he fac ha Romania was characerized by higher han usual growh raes during he analyzed period, a leas growh raes higher han he long-run growh rae of he developed economies. The parameer characerizing he money supply growh mean, m *, is esimaed a a value close o he parameer mean. 4. Forecasing GDP in he long run In his secion, I made a few forecass regarding he economic growh of Romania by using he esimaed model. The firs forecasing exercise projeced he quarerly GDP for he period beween quarer 1-28 and quarer Alhough he quarerly models are used mos of he ime o forecas he GDP for 8 o 1 quarers, he fac ha his model includes he rend suggess he possibiliy of using his model o forecas he long run endencies of he Romanian economy. A he same ime, since his model is a srucural one, one can derive a characerizaion of he long-run behavior of he economy based on he deep parameers. I also compued a confidence inerval given by considering he uncerainy wih respec o boh shocks and parameers. Figure 1 shows a seady endency o grow for he horizon. A he same ime, he confidence inerval showed a cerain uncerainy wih respec o he growh rae. Figure 1 Forecasing quarerly GDP for Source: Auhor s own compuaions. 8 Romanian Journal of Economic Forecasing 3/29

7 Forecasing he Romanian GDP The above graph can be beer undersood when aking ino consideraion he annualized growh rae of GDP, as shown in Figure 2. The forecas was done for a mean average growh rae for he horizon. We can see a higher growh poenial for he firs years. This esimaion of he growh poenial of over 5.3% in he period indicaes ha, in he medium run, he Romanian economy can grow a a high pace. A he same ime, given he values of he srucural parameers in he model, in he long run he economy ends oward a growh poenial of abou 4.9%. Obviously, possible modificaions in he srucural parameers can lead o smaller or higher long-run growh raes. Figure 2 Forecasing annualized growh rae for quarerly GDP Source: Auhor s own compuaions. In Figure 3, I compare he long-run forecass of he esimaed DSGE model wih he long-run forecas of he calibraed model in Solow, as shown in Caraiani (27a). We can again see again ha he forecased growh rae sabilizes around 4.9% saring in 212. A he same ime, according o he underlying hypohesis, he Solow model predics a decreasing growh rae which ends, in he long run, o he TFP growh rae, in he presen case, 3%. I consider ha he forecas of he DSGE model is more credible under he circumsances ha I could esimae he srucural parameers. Also, i mus be said ha he forecas of he DSGE model allows for he esimaion of a confidence inerval, a hing no possible wih he calibraed Solow model, which also allows deviaions from he forecased mean. Romanian Journal of Economic Forecasing 3/29 81

8 Insiue of Economic Forecasing Figure 3 Forecasing long-run economic growh using he DSGE model and he Solow model Source: Auhor s own compuaions. The resuls from he DSGE forecas are also close o he official forecass from he Naional Commission for Forecasing (28). The Commission s forecass are a bi higher for he period, bu in he long run, owards he 22 horizon, he differences become small, being higher by less han.4%. I mus also be said ha alhough he Commission s forecass are higher han he mean forecas of he DSGE model, hey are wihin he compued confidence inerval. Conclusion One of he mos imporan applicaions of a macroeconomic model is forecasing. A he same ime, he qualiy of he forecasing is one of he mos significan ess regarding he relevance of a cerain macroeconomic model. In his paper, I esimaed a moneary DSGE model which I used in order o forecas he GDP in he long run. This paricular DSGE model has wo feaures ha differeniae i from oher DSGE models applied o Romania. Namely, i is characerized by a banking secor, and also includes he presence of rends in he observables variables, GDP and inflaion. The presence of hese feaures led o paricular resuls for he forecass of GDP. Thus, for he horizon, he model confirms he presen consensus beween economiss regarding a growh poenial of Romania of 5 o 6%. In he long run, he model predics a growh rae of over 4.9%. Due o changes in he srucural parameers, which can be influenced by economic policies, or by he behavior of he economic agens, or changes in he rend, i is possible ha he acual growh rae be significanly differen from he forecased one. 82 Romanian Journal of Economic Forecasing 3/29

9 Forecasing he Romanian GDP References Albu, L.L. and A. Roudoi, (23), Facor and Mechanisms of Economic growh in Transiional Economies of Differen Types (Case of Romania), Romanian Journal of Economic Forecasing, 5(4): An, S. and F. Schorfheide, (27), Bayesian Analysis of DSGE Models. Economeric Reviews, 26 (2-3): 1-6. Brooks, S.P., and A. Gelman, (1998), General Mehods for Monioring Convergence of Ieraive Simulaions, Journal of Compuaional and Graphical Saisics, 7(4): Caraiani, P. (27a), Modeling he Economic Growh in Romania wih he Solow Model, Romanian Journal of Economic Forecasing, 8(1): Caraiani, P. (27b), An Analysis of Economic Flucuaions in Romanian Economy Using he Real Business Cycle Approach. Romanian Journal of Economic Forecasing, 8(2): Caraiani, P. (28), Forecasing Romanian GDP Using a Small DSGE Model, Romanian Journal of Economic Forecasing, 9(1): Chari, V.V., L. Jones and R. Manuelli. (1995), The Growh Effecs of Moneary Policy, Federal Reserve Bank of Minneapolis Quarerly Review, 19(4): Cogley, T. and J.M. Nason. (1994), "Tesing he Implicaions of Long-Run Neuraliy for Moneary Business Cycle Models," Journal of Applied Economerics, 9(S):37-7. Cooley, T.F and G.D. Hansen, Gary D, (1989), "The Inflaion Tax in a Real Business Cycle Model," American Economic Review, 79(4): Croioru L. and T. Cornel. (23), The Romanian Growh Poenial A CGE Analysis, Romanian Journal of Economic Forecasing, 4(4):7-23; Dobrescu, E. (26), Macromodels of he Romanian Marke Economy. Ediura Economică, Buchares. Kydland, F. and Ed. Presco. (1982), Time o Build and Aggregae Flucuaions. Economerica 5 (6): Hansen, G. (1985), Indivisible Labor and he Business Cycle. Journal of Moneary Economics, 16(3): Naional Commission for Forecasing. (28), The Projecion of he Main Macroeconomic Indicaors up o 22, The Spring Long Term Forecasing, Buchares, Romania. Schorfheide, F. (2), Loss Funcion-Based Evaluaion of DSGE Models, Journal of Applied Economerics, 15(6): Walsh, Carl. (23), Moneary Theory and Policy. 2 nd Ediion. MIT Press. Romanian Journal of Economic Forecasing 3/29 83

10 Insiue of Economic Forecasing Appendix Bayesian Esimaion Resuls SE_e_a SE_e_m gam ms rho psi Romanian Journal of Economic Forecasing 3/29