THE USE OF QUALITATIVE INFORMATION FOR FORECASTING EXPORTS *

Transcription

1 Aricles Winer 2006 THE USE OF QUALITATIVE INFORMATION FOR FORECASTING EXPORTS * Fáima Cardoso** Cláudia Duare** 1. INTRODUCTION The analysis of he evoluion of exernal rade, in paricular of expors, is very imporan in he case of small open economies, such as he Poruguese one. However, he release of exernal rade daa, by Insiuo Nacional de Esaísica (INE) 1 only happens abou 10 weeks afer he end of he reference quarer. Due o his ime lag, shor-run expors forecass play an imporan role in monioring economic aciviy. Neverheless, forecasing expors is no sraighforward. Owing o he volailiy of his series, forecasing expors enails a paricularly high uncerainy level. Moreover, calculaing imely forecass requires he idenificaion of variables ha no only may bring useful informaion for forecasing expors, bu are also imely released. Survey daa can be, in his conex, paricularly helpful for forecasing expors. According o an increasing lieraure on he use of qualiaive daa for forecasing quaniaive aggregaes, surveys are a very imporan ool for shor-run economic analysis and forecasing. For example, Mourougane and Roma (2002), and Mourougane (2006) find evidence ha confidence indicaors are useful for forecasing, in he shor run, he real rae of change of GDP, of some euro area counries (in he firs case), and of Canada (in he second case). More recenly, Cuenca e Millaruelo (2006) analyse he relevance of considering a se of qualiaive indicaors for assessing he behaviour of he main macroeconomic aggregaes of he euro area. The use of survey daa has drawn more and more aenion mainly because hese daa are released wih shor lags and are no revised. Furhermore, in he case of quesions abou agens expecaions for he near fuure, survey series may be seen as leading indicaors. The aim of his aricle is o assess he role of qualiaive informaion for forecasing he year-on-year rae of change of goods expors, in nominal erms. 2 In paricular, for obaining shor-run expors forecass, we esimae models in which only survey daa are exploied (released by he European Commission) referring o he Manufacuring Indusry Survey and o he Economic Senimen Indicaor. The aricle is organized as follows. Secion 2 describes he daa. Secion 3 presens he selecion of an economeric model for he rae of change of Poruguese expors, in nominal erms, based on qualiaive indicaors, being is performance evaluaed hrough an ou-of-sample forecasing exercise. Finally, secion 4 concludes. * The opinions and findings expressed in his aricle are hose of he auhors and do no necessarily coincide wih hose of he Banco de Porugal. The auhors hank Anónio Rua and Paulo Eseves for heir suggesions and commens. The usual disclaimer applies. ** Economic Research Deparmen. (1) Porugal s Naional Saisical Insiue. (2) Iniially boh he nominal and he real rae of change were considered. However, furher analysis showed ha he resuls for he real rae of change were significanly worse han hose for he nominal rae of change. This evidence suggess ha, in he cases considered, he enrepreneurs assessmen, underlying heir answers o he survey, is more influenced by nominal measures han by volume measures. Therefore, in his aricle are only presened he resuls for he year-on-year rae of change of expors, in nominal erms. Economic Bullein Banco de Porugal 67

2 Winer 2006 Aricles 2. DATA The daabase used consiss of quarerly series covering he period from he firs quarer of 1996 o he second quarer of The purpose of he analysis is o forecas he year-on-year rae of change of goods expors, in nominal erms, released by INE. As i can be seen in Char 1, his variable denoes some volailiy. For forecasing he developmens in expors qualiaive series are used, boh monhly and quarerly, referring o he Manufacuring Indusry Survey, released by he European Commission. This enquiry, wihin he scope of he harmonised European enquiries, aims a collecing informaion abou enrepreneurs opinion regarding he evoluion of heir companies aciviy. The survey, monhly delivered o a sample of abou 1117 companies, consiss of 18 quesions (8 monhly and 10 quarerly). These quesions can refer o ex-ane (nex hree monhs) or o ex-pos periods (reference monh or las hree monhs). Considering he definiion of he answer collecing period, he evaluaion of he reference monh is based on he firs 20 days of ha monh. From he se of 18 quesions, i was made an iniial selecion, being only chosen he quesions ha are direcly relaed o expors. In paricular, he analysis encompasses four quesions from he quarerly survey and anoher four quesions from he monhly survey. In he case of he quarerly survey, he quesions chosen refer o new orders recenly received (nenc ), o expor expecaions for he nex monhs (eexp ) and o he compeiive posiion in he inra- and exra-communiy markes (pci e pce ). Combining hese las wo variables, and using as weighs he raio of inra- and exra-communiy rade o he oal exernal rade, a new variable was buil (comp ) which can be inerpreed as a global indicaor of Poruguese expors compeiiveness. Regarding he monhly survey, he quesions analysed refer o he recen producion rend (prod ),o oal and expor order books (cenc e cexp ) and o producion expecaions for he nex monhs (eprod ). The qualiaive series are presened as balances, defined as he difference beween posiive and negaive answers, and are seasonally adjused. In addiion o he aggregaed oals, he variables from Char 1 YEAR-ON-YEAR RATE OF CHANGE OF EXPORTS, IN NOMINAL TERMS Per cen Q Q Q Q1 2004Q Q1 Source: INE (Inernaional Trade Saisics). 68 Banco de Porugal Economic Bullein

3 Aricles Winer 2006 he Manufacuring Indusry Survey are also available disaggregaed by economic aciviies (NACE). 3 The aggregaion of he series for each quesion reflecs he weighs of each branch (measured by he urnover) in he oal. Since, in his case, he aim is modelling and forecasing he year-on-year rae of change of expors, insead of only using he original overall series, i was also considered a se of series resuling from aggregaing he disaggregaes by economic aciviy, using as weighs he raios of he expors of each branch o he oal expors (henceforh denoed by he suffix p). Furhermore, adding o he variables from he Manufacuring Indusry Survey, i was buil a new variable, based on he Economic Senimen Indicaors of he several counries of he European Union (EU) (pise ). This variable, which can be seen as a proxy of he exernal demand of Poruguese expors, is a weighed average of he Economic Senimen Indicaors of he several Members of he EU, whose weighs reflec he srucure of Poruguese expors disaggregaed by desinaion counry. In he conex of a quarerly daabase, monhly series have o be ransformed ino quarerly series. In his case, he quarerly figures were obained by averaging he monhly ones, considering more han one way of disribuing he monhs among he quarers. Tradiionally, he firs quarer refers o he monhs of January, February and March, and so on so forh. Moreover, we also considered lags of one monh (henceforh denoed by he suffix d1), corresponding he firs quarer o December, January and February, and of wo monhs (henceforh denoed by he suffix d2), wih he firs quarer equal o he average of November, December and January. This procedure allows esing he exisence of several kinds of lags (in monhs) in he relaion beween he qualiaive variables and expors. 3. MODELLING 3.1. Firs approach: cross correlaions analysis Considering he size of he daabase, in a firs approach, bivariae correlaions beween he variaion of expors and poenial explanaory variables were analysed. From he resuls aained, i is possible o conclude ha he correlaions calculaed from he weighed series, whose weighs reflec he disaggregaion by NACE of Poruguese expors, are, in general, higher han he correlaions from he original daa. Taking ino accoun his evidence (subsequenly confirmed by robusness checks during model esimaion) i was chosen o work wih he weighed series, whose weighs are based on he Poruguese expors srucure. Analysing he correlaions calculaed from hese series, i is possible o idenify hree differen siuaions: in he case of expor expecaions for he nex monhs (eexpp ) he highes correlaion is regisered in a lagged period (-2); for he variables referring o new orders recenly received (nencp ),o expor order books, conemporaneous and lagged one monh (cexpp e cexpd1p ) and o he proxy of exernal demand (pise ) he highes correlaion values are regisered in he conemporaneous period; in all he oher cases he highes value is mos commonly regisered in he period +1 (see Table 1). This evidence suggess ha he variables nencp, eexpp, cexpp, cexpd1p e pise may be seen as coinciden indicaors (and lead indicaor, in he case of eexpp ) of he rae of change of expors. Neverheless, even in he cases where he highes correlaion is regisered in periods afer, he value of he correlaions in and in lagged periods is sill reasonably high. In addiion, he analysis of he able suggess ha using he series from he monhly survey lagged one and wo monhs does no seem o significanly improve correlaions. In fac, he use of lags of less han (3) NACE Saisical classificaion of economic aciviies in he European Communiy. Noe ha, a leas a he disaggregaion level considered in his aricle (wo-digi), here is a direc correspondence beween his classificaion and he curren Poruguese classificaion of economic aciviies (CAE). Economic Bullein Banco de Porugal 69

4 Winer 2006 Aricles Table 1 CORRELATION COEFFICIENTS Manufacuring Indusry Survey Quarerly survey New orders recenly received (nencp) Expor expecaions for he nex monhs (eexpp) Compeiive posiion in he inra-communiy marke (pci) Compeiive posiion in he exra-communiy marke (pce) Global indicaor of compeiiveness (comp) Monhly survey Recen producion rend (prod) Toal order books (cenc) Expor order books (cexp) Producion expecaions for he nex monhs (eprod) Lagged one monh Recen producion rend (prod) Toal order books (cenc) Expor order books (cexp) Producion expecaions for he nex monhs (eprod) Lagged wo monhs Recen producion rend (prod) Toal order books (cenc) Expor order books (cexp) Producion expecaions for he nex monhs (eprod) Economic Senimen Indicaors Proxy of he exernal demand (pise) Lagged one monh (pised1) Lagged wo monhs (pised2) Noe: All correlaions are calculaed for he weighed series, whose weighs reflec he Poruguese expors srucure. The values refer o he correlaion beween he rae of change of expors in he period and he qualiaive indicaors in period i i 4,..., 4. The shaded area denoes he highes correlaion coefficien. 70 Banco de Porugal Economic Bullein

5 Aricles Winer 2006 a quarer (one and wo monhs) does no, in general, increase he correlaion wih he variaion of expors, for lagged and conemporaneous periods Model esimaion The aim of his secion is o choose he model specificaion, namely he variables o include, which allows aaining he bes resuls in erms of forecasing he variaion of expors. This choice akes ino accoun he principle of parsimony, bu, even so, we ry o cover an informaion se as wide as possible. To represen he rae of change of expors, Auoregressive Disribued Lags (ADL) models are considered, wih he following form: k i i i 1, 1,..., L y L x k n in which y is he year-on-year rae of change of Poruguese expors, in nominal erms, x i, are he qualiaive indicaors, L and i L are lag polynomials and is he residual variable ( 0 and sandard deviaion 2 ). For model idenificaion purposes i was used a sample from 1997Q1 o 2004Q2. Due o he ineres in obaining imely forecass, during model specificaion he qualiaive variables were only inroduced conemporaneously and wih lags (n = 4). Furhermore, adding o economic inerpreaion, model selecion was based on he Schwarz crierion. Based on his crierion, he model chosen, 4 as follows, 5 y y 019. eexpp 02. 2cexpp 034. pise (1) Adjused R 2 = 0.79 Dependen variable sandard deviaion = 6.33 Residuals sandard deviaion = 2.88 has as regressors: he dependen variable (rae of change of expors) lagged four periods y 4, which enables he model o capure base effecs; from he quarerly survey, expor expecaions for he nex monhs, lagged wo periods (eexpp -2 ); from he monhly survey, expor order books (cexpp ); and, lasly, he proxy of exernal demand of Poruguese expors (pise ). There is no evidence of auocorrelaion unil he order 4 in his model. Considering he sample ha was no used in model idenificaion (2004Q3 o 2006Q2) i was simulaed an ou-of-sample forecasing exercise. The resuls of his exercise were evaluaed according o he roo mean squared forecas error (RMSFE), which, in his case, is 2.60 (see Char 2). (4) I is worh noing ha oher models, wih similar resuls, were obained. However, he model chosen presened he bes fi in-sample and he bes forecasing performance. (5) The -raios are presened in square brackes. Economic Bullein Banco de Porugal 71

6 Winer 2006 Aricles Char 2 YEAR-ON-YEAR RATE OF CHANGE OF EXPORTS, IN NOMINAL TERMS Observed values Fied values Forecass 10 Per cen Q1 1997Q3 1999Q1 2000Q3 2002Q1 2003Q3 2005Q1 Noe: The shaded area corresponds o a 95% confidence inerval. Comparing he model chosen wih a univariae benchmark, 6 one concludes ha he model wih qualiaive variables has a beer performance, boh in-sample and ou-of-sample, han he univariae model, which presens a residuals sandard deviaion equal o 4.59 and a RMSFE of In order o assess he robusness of he model o he sample used i was carried ou a sensiiviy analysis. According o he resuls of his analysis, he model appears o be robus o changes in he sample dimension. In paricular, when esimaing he seleced model using he full sample (1997Q1 o 2006Q2) he coefficiens remain sable and saisically significan. Sample: 1997Q1 2006Q2 y y 017. eexpp 02. 4cexpp 033. pise (2) Adjused R 2 = 0.77 Residuals sandard deviaion = More imely esimaes: indicaors wih incomplee quarerly informaion The seleced model, described in he previous secion, presens he bes forecasing performance, and gives one esimae o he rae of change of expors abou 70 days before he release of he official daa. (6) The univariae benchmark has he following specificaion: y.. y. y wih -raios presened in square brackes Banco de Porugal Economic Bullein

7 Aricles Winer 2006 In an aemp o obain an esimae even earlier, alernaive models were esimaed in which, saring from he previously seleced model (see (1)), he conemporaneous regressors are replaced wih equivalen series ha resul from only considering informaion abou a fracion of he quarer. Obviously, his is possible because, in he model chosen, he conemporaneous variables are from he monhly survey. Therefore, for boh hese variables (expor order books (cexpp ) and proxy of exernal demand (pise )), alernaive series were calculaed, where he quarerly observaions are replaced by he daa referring o he firs monh (henceforh denoed by he suffix m1) or o he average of he firs wo monhs of he respecive quarer (henceforh denoed by he suffix m2). Consequenly, i is possible o obain a firs esimae of he quarerly variaion of expors from a model wih a specificaion similar o (1) bu in which he variables referring o he expor book orders (cexpp ) and o he proxy of exernal demand of he Poruguese expors (pise ) are replaced by he corresponding variables wih informaion on he firs monh of each quarer only (cexpm1p e pisem1 ). y y 018. eexpp 02. 3cexpm1p 036. pisem (3) In he same way, i can be obained a second esimae using he variables wih informaion on firs wo monhs of each quarer (cexpm2p e pisem2 ). y y 018. eexpp 02. 5cexpm2p 033. pisem (4) Considering as a hird esimae he one obained from he model iniially chosen, ha is, from he model in which all indicaors refer o complee quarers, Table 2 summarises he esimaion and forecasing resuls over he hree sages: As expeced, he resuls obained, boh for he RMSFE and for he accuracy on predicing direcion changes, worsen as less informaion concerning each quarer is used. Neverheless, by incurring his sligh loss of qualiy i is possible o obain more imely esimaes. Furhermore, even considering he model wih informaion only for he firs monh of he quarer, is resuls are sill beer han he ones of he univariae model. The resuls obained sugges ha i is possible o have more imely forecass for he variaion of expors, wihou is qualiy being significanly affeced, by using incomplee quarerly informaion. Moreover, including hese variables in he model allows aking ino accoun he monhly daa as soon as hey are released, wihou having o wai for gahering informaion for he whole quarer. Table 2 MODEL COMPARISON Models RMSFE Maching of he variaion sign (a) (%) Forecas imeliness (b) (days) One monh informaion Two monhs informaion Complee quarer informaion Noes: (a) The maching of he variaion sign measures he number of imes ha he sign of he variaion of he series of ineres (which is he rae of change of expors) is equal o he one of he variaion of he fied values and forecass of he model (in percenage).(b) Forecas imeliness refers o he period of ime elapsed beween he calculaion of he forecas and he release of expors daa. Economic Bullein Banco de Porugal 73

8 Winer 2006 Aricles 4. FINAL REMARKS In his aricle i is assessed he imporance of using qualiaive daa, from opinion surveys, for forecasing expors in he shor run. In paricular, a model for he quarerly year-on-year rae of change of expors is esimaed, considering qualiaive series from he Manufacuring Indusry Survey, referring o Porugal, and he Economic Senimen Indicaors, referring o he EU counries, boh released by he European Commission. Based on he model seleced, some alernaive models are also esimaed in an aemp o obain esimaes as early as possible, vis-à-vis he release of he daa. The models presened produce esimaes for he evoluion of expors beween 70 and 130 days before he release of he expors figures. For he purpose of obaining imely forecass for he variaion of expors, one concludes ha he qualiaive informaion is paricularly relevan for his maer and ha i provides a leading indicaion on expors behaviour. Furhermore, he resuls obained poin o he sabiliy of he model in relaion o he sample used. I should be noed, however, ha forecasing expors enails some difficulies due o he volailiy of he series. The favourable resuls ha were achieved, in his case, can be seen as an incenive o furher research on he use of his mehodology for forecasing oher quaniaive aggregaes. BIBLIOGRAPHY Arís, M. e al. (2003), Forecasing models currenly applied o indicaors compued on he basis of surveys resuls, European Commission. European Cenral Bank (2001), The informaion conen of composie indicaors of he euro area business cycle, Monhly Bullein, November. Brion, E. e Culer, J. (1999), The Bank s use of survey daa, Quarerly Bullein, May, Bank of England. European Commission (2006), Using facor models o consruc composie indicaors from BCS daa a comparison wih European Commission confidence indicaors Cuenca, J. e Millaruelo, A. (2006), Las encuesas de opinión y el análisis de coyunura de la acividad real de la UEM, Bolein Económico, January, Banco de España. Hansson, J. e al. (2003), Business Survey Daa: Do They Help in Forecasing he Macro Economy?, The Naional Insiue of Economic Research, Working Paper nº 84. Mourogane, A. e Roma, M. (2002), Can confidence indicaors be useful o predic shor erm real GDP growh?, Working paper nº 133, European Cenral Bank. Mourougane, A. (2006), Forecasing monhly GDP for Canada, Economics Deparmen Working paper nº515, OCDE. Rua, A. (2002), Composie indicaors for he Euro Area economic aciviy, Economic Bullein, Sepember, Banco de Porugal. Sanero, T. e Weserlund, N. (1996), Confidence indicaors and heir relaionship o changes in economic aciviy, Economics Deparmen Working paper nº170, OCDE. Sanos, R. (2003), The use of qualiaive daa for shor erm analysis, Economic Bullein, Sepember, Banco de Porugal. 74 Banco de Porugal Economic Bullein