TAX SMOOTHING: TESTS ON INDONESIAN DATA

Transcription

1 TAX SMOOTHING: TESTS ON INDONESIAN DATA Rudi Kurniawan Deparmen of Economics Macquarie Universiy Macquarie Park, Sydney NSW 3 Ausralia rudi.kurniawan@mq.edu.au Absrac This paper conribues o he lieraure of public deb managemen by esing for ax smoohing behaviour in Indonesia. Tax smoohing means ha he governmen smoohs he ax rae across all fuure ime periods o minimize he disorionary coss of axaion over ime for a given pah of governmen spending. In a sochasic economy wih an incomplee bond marke, ax smoohing implies ha he ax rae approximaes a random walk and changes in he ax rae are nearly unpredicable. For ha purpose, wo ess were performed. Firs, random walk behaviour of he ax rae was examined by underaking uni roo ess. The null hypohesis of uni roo canno be rejeced, indicaing ha he ax rae is nonsaionary and, hence, i follows a random walk. Second, he predicabiliy of he ax rae was examined by regressing changes in he ax rae on is own lagged values and also on lagged values of changes in he govermen expendiure raio, and growh of real oupu. They are found o be no significan in predicing changes in he ax rae. Taken ogeher, he presen evidence seems o be consisen wih he ax smoohing, herefore provides suppor o his heory. Key Words: Tax smoohing, ax rae, public deb managemen, Indonesia JEL Classificaion: H - Efficiency; Opimal Taxaion. INTRODUCTION This paper ess for ax smoohing behaviour in Indonesia. In heir seminal conribuions, Kydland and Presco (977) and Barro (979, 98, and 98) argue ha ax smoohing is boh a posiive and normaive heory of public deb managemen. Tax smoohing means ha he governmen smoohs ax rae across 87

2 all fuure ime periods o minimize he excess burden of axaion for a given pah of governmen spending. Temporary changes in governmen spending and oupu will resul in budge imbalances (surplus or defici), bu here will be no change in he ax rae. To smooh he ax rae, public deb levels are permied o vary over ime. In a sochasic economy wih an incomplee bond marke, his means ha he ax rae approximaes a random walk and, hence, changes in he ax rae are nearly unpredicable. We examine wo implicaions of ax smoohing hypohesis for he case of Indonesia. Firs, we examine he random walk implicaion for he ax rae by esing he uni roo in he ax rae. For his purpose, we use wo mehods of uni roo es; he Augmened Dickey-Fuller (ADF) es and he Phillip-Perron (PP) es. We find ha he ax rae is nonsaionary, consisen wih a random walk. This resul is line wih, among ohers, Barro (979) for US and Kingson (986) for Ausralia. Second, we es wheher changes in ax rae are predicable. For his purpose, changes in he ax rae are regressed on a vecor of lagged variables, including changes in governmen expendiure and growh of real oupu. The resuls reveal ha he ax rae is unpredicable. This suggess ax smoohing by he governmen.. LITERATURE REVIEW There have been many aemps o es he implicaion of he ax smoohing hypohesis by using differen approaches, and he resuls are relaively mixed. Some sudies es he hypohesis by checking ime series behaviour of he ax rae; wheher he ax rae follows a random walk and wheher changes in he ax rae can be prediced. For insance, Barro (98) examines he ime series behaviour of average ax rae in he Unied Saes over he period of He finds ha boh U.S. federal and oal governmen average ax rae do follow a random walk. There is lile explanaory power for ax-rae changes from changes in governmen spending and growh of real oupu. In conras, Sahasakul (986) finds evidence agains he ax smoohing hypoheses for Unied Saes. He ess he hypohesis for he Unied Saes beween by relaing he curren marginal ax rae o his measures of permanen and ransiory componens of he governmen expendiure rae and some oher variables. He finds evidence ha ax rae responds significanly no only o he permanen governmen expendiure rae bu also o he ransiory componen and he general price level, and a ime rend. 88

3 Kingson and Layon (986) es he ax smoohing hypohesis for Ausralia beween 949/5 98/8. They use similar approach as in Barro (98), bu hey build a ax smoohing model ha combines simpliciy, explici microeconomic underpinnings, and applicabiliy o an open economy. They find ha marginal rae of income ax on average male weekly earnings is a random walk, and ha is changes canno be prediced by a vecor of forcing variables which includes changes in governmen spending and growh of real oupu. These resuls confirm ax smoohing hypohesis. Moreover, since hey also some evidence agains random-walk behaviour of average ax raes, public secor oulays and receips as percenage of GDP, and average male weekly earnings, hey ha heir failure o rejec he ax smoohing hypohesis is probably no jus a saisical arifac. More recen sudies use a differen approach o es he ax smoohing hypohesis. Raher han examining he random walk behaviour and he predicabiliy of ax rae, hey es he hypohesis by examining wheher he fiscal defici is informaive abou fuure changes in governmen expendiures. The governmen ses he budge surplus equal o expeced changes in governmen expendiure. When expendiure is expeced o increase, he governmen runs a budge surplus, and when expendiure is expeced o fall, he governmen runs a budge defici. For examples, Huang and Lin (993) and Ghosh (995) examine he ime series properies of Norh American daa using vecor auoregression echnique, and find ha increases in he budge surplus signal fuure increases in governmen expendiure, which is evidence in favour of ax smoohing. In conras, Olekalns (997) finds for Ausralian pos-world War II daa ha he budge surplus has been oo volaile o be fully consisen wih ax smoohing. Olekalns and Crosby (998) examine long-run daa, covering all of he wenieh and some of he nineeenh cenuries for Ausralia, he Unied Kingdom, and he Unied Saes. They found ha ax smoohing is acceped only for he Unied Saes. 3. THEORETICAL FRAMEWORK The model used in his paper is based on Kingson and Layon (983, 986). In he domesic economy, he represenaive privae domesic agen chooses values of consumpion c, labour n, and asse a ha maximize welfare subjec o resource consrain: 89

4 subjec o ( c ) n Max () h = + + ( τ ) θ n + ( + r ) a ( c + a ) = () Asses are claims on he domesic governmen and/or on foreign eniies wih real ineres rae of r. Using λ as he muliplier for he consrains (), he firs-ordernecessary condiions for inerior opimum are: λ = (3) n = θ τ ) (4) ( r = (5) From equaion () and (5) i follows ha oupu is posiively relaed o produciviy and negaively o axes y = θ ( τ ) (6) Lifeime privae consumpion is given by inheried asses plus lifeime disposable income: = c = a + [ ( τ ) + h ] = θ (7) The governmen seeks benevolen finance of is exogenous purchases of goods and service, g, plus exogenous ransfer h. Alhough deb finance is available, lump-sum axaion is no, so ha ax policy mus sele for second bes. Beginning-of-period public deb, b, is such ha iniial deb, b, is predeermined and erminal deb, b, is zero. Holders are domesic and/or foreign residens. Public and privae deb insrumens are perfec subsiues, so he governmen borrows a he prevailing real ineres rae of zero. The governmen s problem is o choose values of τ and b ha solve [ ] Max a + θ ( τ ) + h (8) = 9

5 subjec o τ θ τ ) + b ( b + g + h ) = (9) ( + According o equaion (), he governmen s objecive funcion is he privae agen s indirec lifeime uiliy funcion. Le μ be he mulipliers o he consrains (). The firs-order necessary condiions for an inerior opimum are μ τ = + () τ μ = () μ From () and () we ge τ = τ [ = *] () τ According o (), he ax rae in curren period is an unbiased predicor of fuure ax raes, in oher words, he ax rae is expeced o remain consan. This implies ha he ax rae behaves as a random walk. The random walk condiion for he ax rae series can be wrien in an empirical form as follows: Δ τ = ε, ε ~(, σ ) (3) where Δ τ = τ τ and ε is a whie noise error erm ha is independen and idenically disribued. The random walk hypohesis implies ha none of he auocorrelaions in he residual from he firs differenced ax rae is significanly differen from zero. Noed ha under he null hypohesis of random walk, given τ, no oher lagged informaion should be helpful o predic τ. Accordingly, evidence of lagged informaion ha can predic Δ τ would rejec ax smoohing (Barro, 98). 4. DATA AND METHODOLOGY Daa on he ax rae ( τ ), governmen expendiure rae ( Δ g ), and real oupu growh ( Δ y ) are examined for he period of The ax rae and governmen expendiure rae are respecively calculaed as cenral governmen ax revenue and expendiure divided by Gross Domesic Produc (GDP). The real oupu growh is he growh of real GDP, ha is nominal GDP adjused by 9

6 Consumer Price Index (CPI) wih year as he base year. Daa are drawn from he Economic Key Indicaors published on-line by he Asian Developmen Bank, IMF s Governmen Financial Saisics and Minisry of Finance, Indonesia. We examine he random walk implicaion of ax smoohing by esing he null hypohesis of a uni roo in he ax rae variable. Two uni roo ess are used: he Augmened Dickey-Fuller (ADF) es and he Phillips-Peron (PP) ess. Based on model, changes in he ax rae canno be prediced eiher by is own lagged values or by lagged values of oher variables; he raio of governmen expendiure Δ g and growh of real oupu θ. To begin wih, we es wheher he ax rae can be prediced by is own lagged values by esimaing he following auoregressive equaion: k Δ τ = α + αiδ τ i + u i= (4) The es is carried ou by employing F es wih null hypohesis α = α =... = α k =. The opimal lag lengh is deermined by using he AIC model selecion crierion (Sock, 994). To furher examine he null hypohesis of ax smoohing, changes in he ax rae variable is regressed on lagged values of iself ogeher wih lagged values of changes in governmen expendiure raio Δ g and he growh of real oupu θ. Therefore, he following vecor auoregression (VAR) model is esimaed: Y = δ +Φ Y +Φ Y +... Φ py p + u (5) where Y [ τ g θ ] = Δ, Δ, ' is a vecor of endogenous variables. The Φ i (i=,,..., p) are k-dimensional quadraic coefficiens marices, and u represens he k- dimensional vecor of residual. Analysis is based on he F-saisics and Chisquared of he Wald coefficien ess. As noed by Barro (98), es for he unpredicabiliy of ax rae changes are mos ineresing in an environmen where some fuure changes in real governmen spending, real oupu, ec., are forecasable. I would be less ineresing o find ha ax rae shifs were unpredicable if changes in he governmen spending raio o oupu were also unpredicable. 9

7 5. ESTIMATION AND RESULT Table we presen he resuls of he ADF and PP ess for uni roo in he ax rae variable. All -saisics obained from boh ess, eiher only wih consan or wih consan and rend, are no significan a 5 per cen level (or lower). Therefore, he null hypohesis of a uni roo canno be rejeced. Accordingly, we conclude ha he ax rae saisfies one condiion for a random walk. Table : The ADF and PP Tes Resuls for he Tax Rae Inercep Only Inercep and Trend Uni Roo Tes Lag -sa. p-value Lag -sa. p-value ADF PP Table repors auoregression resuls which provide evidence on wheher changes in ax rae are predicable by is own previous values. A comparison furnished on he basis of Akaike Informaion Crieria (AIC) saisics for differen orders of auoregressive equaions suggess ha he firs order auoregressive model is he bes predicive model. However, raher han jus reporing resuls from model wih one lag, Table also repors resuls from model wih up o four lags so we have richer informaion on predicabiliy of changes in ax rae. The F-saisics obained from all equaions are no significan a 5% or lower. Accordingly, he null of zero coefficiens for all lagged variables canno be rejeced, and we conclude ha changes in ax rae canno be prediced significanly by changes in previous periods. Table : Tax Rae Auoregression: 97 8 Number of Lags in he Model Coefficien Lag 4 Lag 3 Lag Lag α (.5) (.73) (.59) (.363) α (-.4) (-.46) (-.466) (-.46) α (-.45) (-.38) (-.3636) α (.7476) (.583) 93

8 α 4.89 (.58) AIC F-sa Prob. (F-sa.) (.976) (.874) (.855) (.683) Noe: Figures in parenheses are -saisics Table 3 repors he resuls from vecor auoregression (VAR), providing evidence on wheher changes in ax rae are predicable no only by is own pas values bu also by pas changes in governmen expendiure raio and growh of real GDP. Moreover, Table 3 also provides evidence on he predicabiliy of changes in governmen expendiure and oupu growh. The VAR model is esimaed for one up o four lags, and he AIC saisic indicaes ha he model wih wo lags is he bes model specificaion. However, Table 3 also presens he resuls from all lag srucures. The resuls for he ax rae changes Δ τ sugges low predicive power as indicaed by insignifican F-saisics obained from all model a 5% level. Accordingly, he null of zero coefficiens for all lagged variables canno be rejeced. Meanwhile, he resuls from he Wald coefficien resricion ess for individual variables are quie mixed. While he lagged values of changes in he ax rae and governmen expendiure raio are no significan, he lagged values of real oupu growh are significan a 5 per cen level. This indicaes ha changes in ax base, as proxied by oupu growh, migh be imporan for changes in ax rae. However, based on he overall F-saisics, and in accordance wih he resuls from random walk ess, we conclude ha changes in he ax rae are unpredicable by he lagged values for he variables considered in he model. Table 3: Vecor Auoregression Wald Coefficien Resricion Tes Dependen Complee VAR Δ τ Δ g Δ y Variable Lag F-Sa. Prob. Chi-Sqr Prob. Chi-Sqr. Prob. Chi-Sqr. Prob. Δ τ Lag Lag Lag Lag

9 Δ g Lag Lag Lag Lag Δ y Lag Lag Lag Lag VAR AIC Lag =.35 Lag =.875 Lag 3=.9954 Lag 4=.447 Noe: In he Wald es, he coefficiens of lagged variables are resriced o zero. The resuls for changes in governmen expendiure Δ g sugges some predicive power as indicaed by significan F-saisics a 5% level. Therefore, he null of unpredicable changes in governmen expendiure can be rejeced. In paricular, he Wald coefficien ess indicae ha hisorical values of changes in ax rae and governmen expendiure raio have significan predicable effec on changes in governmen expendiure. Meanwhile, he Wald coefficien es saisics for oupu growh are no significan a 5% level, indicaing ha changes in rae of governmen expendiure is no necessarily depends on changes in ax base. Possible explanaion for his is ha he governmen in he maer of fac needs o coninuously increase is expendiure over ime in order o simulae he economy, regardless of he economic condiion. In he case of Indonesia, his was made possible by he suppor of foreign aid and large revenue from oil (during he New Order regime), and, laer on, by rapid increase in ax revenue and larger access o deb marke, boh domesic and inernaional. The governmen, hen, can always finance increasing expendiure wihou heavily relying on he growh of ax base. The resuls for oupu growh Δ y indicae low predicive power. In Table 3, he F-saisics are no significan a 5% level and he null of unpredicable oupu growh canno be rejeced. The Wald coefficien ess confirm ha none of he lagged endogenous variables is significan individually. The oupu growh, hen, is unpredicable eiher by is own hisorical values or by pas values of changes in ax rae and governmen expendiure. Relaed o hese findings, he relaionship beween governmen expendiure and economic growh has acually been a 95

10 coninued debae among scholars. I is possible ha differen componens of expendiure affec real income in differen ways, and vary considerably across he naions. A cross counry sudy by Landau (983) reveals a negaive effec of governmen consumpion expendiure on growh of real oupu. Bose, Haque and Osborn (7) find ha he share of governmen capial expendiure in GDP is posiively and significanly correlaed wih economic growh, bu curren expendiure is insignifican. They also find ha, a he disaggregaed level, governmen invesmen in educaion and oal expendiures in educaion are he only oulays ha are significanly associaed wih growh once. For he case of Indonesia, a number of sudies find ha governmen expendiures do no play a significan role in promoing economic growh (Dogan and Tang 6). A sudy by Ramayandi (3) for he period of even finds ha boh he share of governmen unproducive spending and producive spending show negaive effec on growh. This resul suggess an exisence of inefficiencies in he overall managemen of governmen budge in Indonesia during he period under consideraion. 6. CONCLUSION This paper examined Indonesian ax rae daa o ascerain wheher here is an evidence of ax smoohing. For ha purpose, wo ess were performed. Firs, random walk behaviour of he ax rae was examined by underaking uni roo ess. The null hypohesis of uni roo canno be rejeced, indicaing ha he ax rae is nonsaionary and, hence, i follows a random walk. Second, he predicabiliy of he ax rae is examined by regressing changes in he ax rae on is own lagged values and also lagged changes in he ax rae, changes in he governmen expendiure raio o GDP, and growh of real oupu. They are found o be no significan in predicing changes in he ax rae. Taken ogeher he presen evidence seems o be consisen wih he ax smoohing hypohesis since he ax rae series displays random walk behaviour and is unpredicable. Therefore, he presen empirical sudy provides suppor o his heory. BIBLIOGRAPHY Barro, R. J. (979). On he deerminaion of he public deb, Journal of Poliical Economy 87(5):

11 Barro, R. J. (98), On he predicabiliy of ax-rae changes, NBER Working Paper 636; reprined in Barro, R. (99), Macroeconomic policy, Harvard Universiy Press, Cambridge, MA. Ghosh, A. R. (995), Ineremporal ax-smoohing and he governmen budge surplus: Canada and he Unied Saes, Journal of Money, Credi, and Banking 7, Huang, C.-H. and K. S. Lin (993), Deficis, governmen expendiure, and ax smoohing in he Unied Saes: , Journal of Moneary Economics 3, Kingson, G. H. and A. P. Layon (986). Tax smoohing and Ausralian fiscal policy. Research paper (Macquarie Universiy. School of Economic and FinancialSudies), no. 38. Sahasakul, C. (986), The U.S. evidence on opimal axaion over ime, Journal of Moneary Economics 8,