What Caused IMF’s Fore­cast Er­rors? Anal­y­sis from Po­lit­i­cal Econ­omy Per­spec­tive

XiongAi­zong(熊爱宗)

China Economist - - Contents - Xiong Ai­zong (熊爱宗) In­sti­tute of World Eco­nom­ics and Pol­i­tics, Chi­nese Academy of So­cial Sci­ences, Beijing, China

Ab­stract: Pro­vid­ing high-qual­ity eco­nomic fore­casts is an im­por­tant re­spon­si­bil­ity of the In­ter­na­tional Mon­e­tary Fund (IMF) in main­tain­ing world fi­nan­cial and eco­nomic sta­bil­ity. How­ever, er­rors are in­evitable in IMF eco­nomic fore­casts for its mem­ber coun­tries. Based on fore­cast method and in­for­ma­tion, and po­lit­i­cal fac­tor, this paper cre­ates a po­lit­i­cal eco­nom­ics frame­work for an­a­lyz­ing the IMF’s fore­cast er­rors, and tests the ef­fects of var­i­ous fac­tors on the IMF’s fore­casts us­ing the panel data anal­y­sis method. Ac­cord­ing to our find­ings, if a coun­try re­ceives IMF loans and shares a sim­i­lar vote with the United States at the UN General As­sem­bly, it will more likely re­ceive an op­ti­mistic fore­cast by the IMF. Mean­while, mem­ber coun­tries’ data avail­abil­ity and IMF fore­cast er­rors for ma­jor economies may also af­fect fore­cast on a coun­try. There­fore, this paper pro­poses rec­om­men­da­tions on fur­ther im­prov­ing the IMF’s fore­cast qual­ity by cre­at­ing more in­de­pen­dent fore­cast pro­ce­dures and en­hanc­ing fore­cast data qual­ity and fore­cast ac­cu­racy.

Key­words: In­ter­na­tional Mon­e­tary Fund, sur­veil­lance, eco­nomic fore­cast, po­lit­i­cal

econ­omy

JEL Clas­si­fi­ca­tion Codes: F33; F53; F55

DOI:1 0.19602/j .chi­nae­conomist.2018.11.0619602/ j .chi­nae­conomist.2018.09.02

1. In­tro­duc­tion

Macroe­co­nomic fore­cast­ing is an im­por­tant as­pect of the In­ter­na­tional Mon­e­tary Fund’s (IMF) sur­veil­lance ac­tiv­i­ties. By fore­cast­ing global, re­gional, and mem­ber economies’ de­vel­op­ment trends, the IMF mon­i­tors world eco­nomic op­er­a­tions, iden­ti­fies pos­si­ble sta­bil­ity risks, pro­poses rec­om­men­da­tions, and en­sures world eco­nomic and fi­nan­cial sta­bil­ity. Ac­cu­rate eco­nomic fore­cast­ing is of great im­por­tance both to the IMF and its mem­ber coun­tries. For the IMF it­self, pre­cise eco­nomic fore­cast­ing is an im­por­tant ba­sis for pro­vid­ing pol­icy sup­port. This in­cludes the pro­vi­sion of not only fi­nan­cial re­lief ar­range­ments to mem­ber coun­tries be­fore or dur­ing an eco­nomic cri­sis, but pol­icy rec­om­men­da­tions dur­ing eco­nomic sta­bil­ity as well. For mem­ber coun­tries, the IMF’s eco­nomic fore­cast­ing serves as an im­por­tant ref­er­ence for their plan­ning and de­ci­sion-mak­ing ac­tiv­i­ties. In ad­di­tion, the IMF’s fore­cast has an im­por­tant in­flu­ence on lenders and bor­row­ers, rat­ing agen­cies, the me­dia, and the pub­lic at large. There­fore, the IMF’s eco­nomic fore­cast should be sound, even­handed, and of high qual­ity (IEO, 2014).

But, in re­al­ity, er­rors in the IMF’s eco­nomic fore­casts on its mem­ber coun­tries are in­evitable. Fig­ure 1 il­lus­trates the er­rors of the IMF’s fore­casts on real eco­nomic growth of 163 mem­ber coun­tries from 2003 to 2015. As can be seen from the chart, such er­rors are man­i­fested in over­es­ti­mates (pos­i­tive val­ues) and un­der­es­ti­mates (nega­tive val­ues). Over­all, the IMF un­der­es­ti­mated mem­ber coun­tries’

growth in 2003-2007 and 2010, and over­es­ti­mated their growth in 2008-2011 and 2015.

Er­rors de­rive from de­vi­a­tion of eco­nomic vari­ables’ fore­cast val­ues from real val­ues. The fol­low­ing two meth­ods may be ap­plied in ex­plor­ing what led to fore­cast er­rors, di­rect and in­di­rect. First, the di­rect method seeks to iden­tify the dif­fer­ences be­tween the eco­nomic vari­able fore­cast be­hav­ioral equa­tion and real vari­able op­er­a­tion equa­tion, but this method is un­able to pre­cisely iden­tify a vari­able’s real op­er­a­tion equa­tion. Se­cond, the in­di­rect method seeks to iden­tify fac­tors that may cause the fore­cast be­hav­ioral equa­tion to de­vi­ate from the real op­er­a­tion equa­tion. This paper adopts the in­di­rect method, and cre­ates a po­lit­i­cal econ­omy frame­work for an­a­lyz­ing the IMF’s fore­cast er­rors from three as­pects: fore­cast method, fore­cast in­for­ma­tion, and po­lit­i­cal fac­tor. Based on panel data anal­y­sis method, our con­clu­sion is that the IMF’s fore­cast on a coun­try is in­flu­enced by whether the coun­try re­ceives IMF loans, its po­lit­i­cal re­la­tion­ship with the IMF, its eco­nomic data avail­abil­ity, and the IMF’s fore­cast er­rors for ma­jor economies.

2. Lit­er­a­ture Re­view

Pro­vid­ing high-qual­ity eco­nomic fore­casts is a com­par­a­tive ad­van­tage of in­ter­na­tional eco­nomic or­ga­ni­za­tions re­gard­ing in­for­ma­tion sup­ply (Fra­tianni and Pat­ti­son, 1982). How­ever, most stud­ies find the ex­is­tence of de­vi­a­tions in IMF eco­nomic fore­casts. Judg­ing by the lit­er­a­ture, the IMF’s fore­casts are be­lieved to be in­flu­enced by IMF loans, data avail­abil­ity, ini­tial con­di­tions of fore­cast, po­lit­i­cal fac­tors, and the IMF em­ploy­ees’ fore­cast ca­pa­bil­i­ties (Gen­berg and Martinez, 2014).

First, most stud­ies find that IMF fore­cast er­rors are cor­re­lated to whether a coun­try re­ceives IMF loans. Beach et al. (1999) finds that an in­crease of IMF loans by one bil­lion Spe­cial Draw­ing Right­ings ( SDRs) to coun­tries in the Western Hemi­sphere in­creases fore­cast er­ror by 0.17 per­cent­age points. How­ever, the IMF’s In­de­pen­dent Eval­u­a­tion Of­fice (IEO, 2014) notes that not all

loans in­flu­ence IMF fore­casts. Fore­cast er­rors are rel­a­tively sig­nif­i­cant only for coun­tries that re­ceive IMF loans through “ex­cep­tional ac­cess,” with most such er­rors be­ing re­duced or re­versed at the first pro­gram re­view.

Se­cond, the IMF fore­cast er­rors are cor­re­lated to data avail­abil­ity on whether a coun­try adopts the IMF’s data pub­li­ca­tion stan­dard. The IMF pro­vides bet­ter fore­casts on de­vel­oped coun­tries than on de­vel­op­ing ones. The fore­cast data in loan pro­grams for de­vel­oped coun­tries are more trans­par­ent and less volatile be­cause the data are di­rectly used as the ba­sis of fore­cast (Ar­tis, 1996). Tong (2004) an­a­lyzes whether and how data trans­parency stan­dards af­fect macroe­co­nomic fore­casts. Based on macroe­co­nomic growth quar­terly fore­cast data of 16 coun­tries from 1996 to 2003, this study finds that trans­parency stan­dards are of great in­flu­ence on the im­prove­ment of the IMF’s fore­cast ac­cu­racy. Mrkaic ( 2010) in­ves­ti­gates how mem­ber coun­tries’ par­tic­i­pa­tion in the IMF’s Data Stan­dards Ini­tia­tives ( DSI) may in­flu­ence the fore­cast qual­ity of the World Eco­nomic Out­look. Re­sults show that the World Eco­nomic Out­look’s fore­casts on mem­ber coun­tries that adopted the Spe­cial Data Dis­sem­i­na­tion Stan­dards ( SDDS) were gen­er­ally bet­ter than the fore­casts on those that only adopted the General Data Dis­sem­i­na­tion Stan­dards (GDDS) and those that did not par­tic­i­pate in the DSI at all.

Third, IMF fore­cast er­rors are cor­re­lated to in­suf­fi­cient con­sid­er­a­tion of as­sump­tion of ini­tial con­di­tions and in­ter­na­tional vari­ables. By an­a­lyz­ing the IMF’s macroe­co­nomic fore­casts on var­i­ous re­gions, Tak­agi and Ku­cur (2006) find that un­ex­pected changes in the mon­e­tary poli­cies of ma­jor economies and oil prices sig­nif­i­cantly in­flu­ence the IMF’s fore­casts. Tim­mer­mann (2006) notes that er­rors in the fore­cast of U.S. GDP growth rates are sig­nif­i­cantly pos­i­tively cor­re­lated with er­rors in the fore­casts of GDP growth rates of most de­vel­oped coun­tries in the same years. Fore­cast ac­cu­racy is also re­lated to the as­sump­tion of out­put gaps.

Fourth, IMF fore­cast er­rors are cor­re­lated to the im­pact of po­lit­i­cal fac­tors. Alden­hoff (2007) notes that IMF eco­nomic fore­casts on real eco­nomic growth, in­fla­tion, and un­em­ploy­ment rate are of­ten dis­torted by po­lit­i­cal bias. Dreher et al. (2008) carry out a po­lit­i­cal econ­omy anal­y­sis of the IMF’s fore­cast data on 157 coun­tries dur­ing 1999-2005, and put for­ward a hy­poth­e­sis on po­lit­i­cal strat­egy that points out that the gov­ern­ments of IMF mem­ber coun­tries ex­ert pres­sures on the IMF through pub­lic or con­cealed chan­nels to cause the IMF to make op­ti­mistic eco­nomic fore­casts on their coun­tries. Em­pir­i­cal re­sults re­veal that coun­tries with votes closer to those of the United States at the UN General As­sem­bly re­ceive lower fore­casted in­fla­tion rates from the IMF.

Fifth, IMF fore­cast er­rors are cor­re­lated to IMF em­ploy­ees’ fore­cast­ing ca­pa­bil­i­ties and ex­pe­ri­ence. Gen­berg and Martinez (2014) note that IMF coun­try rep­re­sen­ta­tives have con­sid­er­able au­ton­omy over the fore­casts of the World Eco­nomic Out­look, and so it is par­tic­u­larly de­ci­sive what fore­cast meth­ods they use and how ca­pa­ble and ex­pe­ri­enced they are.

Ex­ist­ing stud­ies in­ves­ti­gate var­i­ous fac­tors’ ef­fects on the ac­cu­racy of IMF fore­casts from dif­fer­ent per­spec­tives, ei­ther sep­a­rately or col­lec­tively. Like the lit­er­a­ture, this paper the­o­ret­i­cally and em­pir­i­cally sheds light on what in­flu­ences the IMF’s fore­casts us­ing a po­lit­i­cal econ­omy frame­work. This paper iden­ti­fies the sources of IMF fore­cast er­rors from three as­pects ( fore­cast method and in­for­ma­tion, and po­lit­i­cal fac­tors), and ex­am­ines dif­fer­ent fac­tors’ ef­fects on IMF fore­cast er­rors us­ing panel data. Un­like the lit­er­a­ture, this paper in­cludes data avail­abil­ity and ini­tial fore­cast con­di­tions into the an­a­lyt­i­cal frame­work. Re­gard­ing data avail­abil­ity, this paper ex­am­ines both the IMF’s Spe­cial Data Dis­sem­i­na­tion Stan­dards ( SDDS) and the World Bank’s Sta­tis­ti­cal Ca­pa­bil­ity In­di­ca­tor (SCI) to ver­ify the ef­fects of in­creased data avail­abil­ity on re­duc­ing fore­cast er­rors. Re­gard­ing ini­tial fore­cast con­di­tions, this paper in­ves­ti­gates how IMF fore­cast er­rors for the United States, Ger­many and China are cor­re­lated with fore­cast er­rors for other coun­tries, and thus ver­i­fies the ex­pec­ta­tion that the IMF’s fore­casts on ma­jor economies will sig­nif­i­cantly in­flu­ence fore­casts on other coun­tries.

3. Sources of IMF Fore­cast Er­rors: An­a­lyt­i­cal Frame­work

Ref­er­enc­ing Atoian et al. (2004) and Sahin (2014), this paper as­sumes that the IMF’s be­hav­ioral equa­tion of macroe­co­nomic fore­cast is as fol­lows:

Ac­cord­ing to equa­tion (3), IMF fore­cast er­rors de­rive from the fol­low­ing sources:

First, fore­cast er­rors are de­rived from the dif­fer­ence be­tween fore­cast model or method and real be­hav­ioral equa­tion , which is re­lated to de­vi­a­tions in the IMF’s fore­cast be­hav­iors. In ad­di­tion, the dif­fer­ence be­tween fore­cast model and real be­hav­ioral equa­tion caused by de­vi­a­tions in fore­cast­ing be­hav­iors may be de­rived from var­i­ous sources.

One source of er­rors is data avail­abil­ity prob­lems. In some cir­cum­stances, data avail­abil­ity di­rectly de­ter­mines the fore­cast model or method em­ployed by the IMF. IEO (2014) notes that data avail­abil­ity is the sin­gle most im­por­tant fac­tor that in­flu­ences coun­try rep­re­sen­ta­tives’ choice of fore­cast mod­els. Due to poor data avail­abil­ity, sta­tis­ti­cal mod­els such as struc­tural eco­mo­met­ric mod­els, vec­tor-au­tore­gres­sion, or re­duced-form equa­tions play a much smaller role in eco­nomic fore­cast­ing of low-in­come coun­tries com­pared with ad­canced coun­tries. Given data re­stric­tions and in­suf­fi­cient ex­ter­nal fore­cast re­sources that can be ref­er­enced, the IMF has to rely on the as­sess­ments of coun­try rep­re­sen­ta­tives in mak­ing fore­casts on low-in­come coun­tries (Gen­berg and Martinez, 2014).

An­other source of er­rors is de­vi­a­tions in fore­cast­ing be­hav­iors aris­ing from the IMF’s re­spon­si­bil­i­ties. Con­sid­er­ing the IMF’s im­por­tance, its World Eco­nomic Out­look has an im­por­tant in­flu­ence on the world econ­omy as a whole and var­i­ous coun­tries in­di­vid­u­ally. The IMF’s fore­cast on the eco­nomic de­vel­op­ment con­di­tions of var­i­ous coun­tries di­rectly in­flu­ences the mar­ket sen­ti­ments and eco­nomic per­for­mance of such coun­tries. An im­por­tant duty of the IMF is to pro­mote global eco­nomic sta­bil­ity. For this rea­son, the IMF has a nat­u­ral in­cli­na­tion to make op­ti­mistic fore­casts to guide mar­ket ex­pec­ta­tions and pro­mote world eco­nomic de­vel­op­ment. Such in­cli­na­tion is par­tic­u­larly ob­vi­ous in an eco­nomic down­turn, when boost­ing eco­nomic ex­pec­ta­tions can pre­vent down­ward risks and the sel­f­re­al­iza­tion of a vi­cious cy­cle. IMF em­ploy­ees, for the most part, tend to play down risks and to avoid li­a­bil­ity for con­ta­gious ef­fect of fi­nan­cial crises by overem­pha­siz­ing risks (Dreher et al., 2008). But such an op­ti­mistic ten­dency does not ap­ply to all coun­tries. Con­sid­er­ing the world econ­omy’s con­ta­gious ef­fect, the IMF’s op­ti­mistic ten­dency is more likely to be man­i­fested in the fore­casts on the world econ­omy as a whole and ma­jor coun­tries and re­gions.

More­over, de­vi­a­tions in the fore­cast­ing be­hav­iors may also be in­duced by IMF loans. Most stud­ies show that the IMF tends to over­es­ti­mate the ef­fec­tive­ness of its lend­ing pol­icy and to pre­serve its rep­u­ta­tion as an in­ter­na­tional fi­nan­cial sta­bi­lizer. The IMF’s coun­try-of­fice em­ploy­ees tend to make op­ti­mistic fore­casts on mem­ber coun­tries re­ceiv­ing IMF loans to jus­tify their loans. The IMF’s op­ti­mistic

fore­casts on re­cip­i­ent coun­tries also help re­cip­i­ent gov­ern­ments to per­suade other do­mes­tic stake­hold­ers to ac­cept IMF loans.

Se­cond, fore­cast er­rors are de­rived from dif­fer­ences be­tween avail­able in­for­ma­tion and real in­for­ma­tion in the fore­cast­ing process. Avail­able in­for­ma­tion em­ployed in the fore­cast­ing process is of­ten in­com­plete, which leads to fore­cast er­rors.

Dif­fer­ences be­tween avail­able in­for­ma­tion dur­ing fore­cast­ing and real in­for­ma­tion may de­rive from three sources: data avail­abil­ity, as­sump­tions on the re­li­a­bil­ity of poli­cies, and po­lit­i­cal fac­tors. First, data avail­abil­ity is the foun­da­tion for macroe­co­nomic fore­cast­ing. In ad­di­tion to in­flu­enc­ing the above­men­tioned fore­cast­ing meth­ods, data avail­abil­ity will also di­rectly in­flu­ence fore­cast qual­ity.

Se­cond, fore­cast­ing a coun­try’s growth re­quires as­sump­tions on the re­li­a­bil­ity of its do­mes­tic poli­cies, such as mon­e­tary and fis­cal pol­icy, and in­ter­na­tional fac­tors like world eco­nomic growth, eco­nomic per­for­mance, in­ter­na­tional fi­nance, trade, and bulk com­mod­ity prices. Er­rors in as­sump­tions on the re­li­a­bil­ity of poli­cies will af­fect the eco­nomic fore­cast on a coun­try. If the IMF fails to cor­rectly fore­cast pol­icy changes of its mem­ber coun­tries, de­vi­a­tions in its fore­cast on such coun­tries will oc­cur.

Third, po­lit­i­cal fac­tor will also cause an im­pact on the fore­cast out­come. Fore­cast­ing eco­nomic vari­ables in­volves both eco­nomic and po­lit­i­cal fac­tors.

Such po­lit­i­cal in­flu­ence may come from two sources: di­rect and in­di­rect. First, mem­ber coun­tries have a di­rect po­lit­i­cal in­flu­ence on the IMF, in­clud­ing their quo­tas and vot­ing rights. IMF em­ploy­ees work un­der the guid­ance of the Ex­ec­u­tive Board, and var­i­ous busi­ness ac­tiv­i­ties have to con­sider the pref­er­ences of ma­jor share­holder coun­tries. There­fore, it is eas­ier for the IMF’s ma­jor share­holder coun­tries to ac­quire more fa­vor­able eco­nomic fore­casts.

Se­cond, mem­ber coun­tries have an in­di­rect po­lit­i­cal in­flu­ence on the IMF. They may seek the IMF’s sup­port in eco­nomic fore­cast­ing through close po­lit­i­cal ties with the IMF’s ma­jor share­hold­ers (such as the United States), which is con­ducive to op­ti­mistic IMF fore­casts on them. Most stud­ies de­ter­mine in­di­rect po­lit­i­cal ties to the IMF by mea­sur­ing the con­sis­tency of a coun­try’s votes with U.S. votes at the UN General As­sem­bly.

Ta­ble 1 iden­ti­fies a few pos­si­ble sources of IMF fore­cast er­rors, as well as pos­si­ble di­rec­tions of var­i­ous fac­tors’ im­pact on such fore­casts. These fac­tors may in­flu­ence dif­fer­ent as­pects of fore­cast­ing. Data avail­abil­ity may in­flu­ence the choice of fore­cast model and method, as well as fore­cast qual­ity. Loan is­suance may cause the IMF to over­es­ti­mate a re­cip­i­ent coun­try’s eco­nomic per­for­mance, but helps IMF em­ploy­ees gain a deeper un­der­stand­ing of such coun­try and in­crease the fore­cast ac­cu­racy on the

1 coun­try as well.

4. Em­pir­i­cal Anal­y­sis

4.1 Model Set­ting

We cre­ate the fol­low­ing model

(1) Fore­cast Er­ror ( ). is the IMF’s fore­cast er­ror for the real GDP growth rate of coun­try i dur­ing pe­riod t , as de­fined in equa­tion (3). Here, and are the IMF’s fore­cast value of real GDP growth rate of coun­try i dur­ing pe­riod t and real GDP growth rate of coun­try i dur­ing pe­riod t . Con­sid­er­ing data avail­abil­ity, this paper in­cludes a to­tal of 160 sam­ple coun­tries, in­clud­ing 129 de­vel­op­ing economies and 31 ad­vanced economies, and sam­ple pe­riod is from 2003 to 2015. Data are from the IMF’s World Eco­nomic Out­look Data­bases.

(2) Eco­nomic size of mem­ber coun­tries ( ). is the share of coun­try i in world GDP dur­ing pe­riod t by cur­rent-price GDP. Data are from the IMF’s World Eco­nomic Out­look Data­bases.

(3) IMF loans ( ). is the amount of IMF loans re­ceived by coun­try i dur­ing pe­riod t as a share in its cur­rent-pe­riod GDP. Data are from the IMF Mon­i­tor­ing of Fund Ar­range­ments Data­base.

(4) Mem­ber coun­tries’ di­rect po­lit­i­cal in­flu­ence on the IMF ( ). Here, we use two in­di­ca­tors to de­pict mem­ber coun­tries’ di­rect po­lit­i­cal in­flu­ence on the IMF ( and ), where is the quota of coun­try i at the IMF dur­ing pe­riod t . is whether per­son­nel from coun­try i serve as ex­ec­u­tive di­rec­tor of the IMF Ex­ec­u­tive Board dur­ing pe­riod t . Where, data are from the IMF Fi­nan­cial Data­base, and data are from the IMF annual re­ports.

(5) Mem­ber coun­tries’ in­di­rect po­lit­i­cal in­flu­ence on the IMF ( ). is the con­sis­tency of coun­try i with the U.S. votes at the UN General As­sem­bly dur­ing pe­riod t . If is 1, all the coun­try’s votes are con­sis­tent with U.S. votes. If it is -1, the coun­try’s votes are to­tally op­po­site to U.S. votes. It is from the United Na­tions General As­sem­bly Vot­ing Data of Voeten et al. (2009).

( 6) Data avail­abil­ity ( ). This paper em­ploys two in­di­ca­tors to mea­sure a coun­try’s data avail­abil­ity. The first in­di­ca­tor is whether a coun­try adopts the IMF’s Spe­cial Data Dis­sem­i­na­tion Stan­dards (SDDS) ( ). If a coun­try adopts the SDDS, is 1; oth­er­wise it is 0. The se­cond in­di­ca­tor is the sta­tis­ti­cal ca­pac­ity in­di­ca­tor of the World Bank in eval­u­at­ing the sta­tis­ti­cal ca­pa­bil­i­ties

of coun­tries (Sta­tis­ti­cal Ca­pac­ity In­di­ca­tor) ( ). This World Bank in­di­ca­tor eval­u­ates coun­tries’ sta­tis­ti­cal ca­pa­bil­i­ties for a score range of 0-100. In­for­ma­tion about the adop­tion of SDDS by coun­tries is from the Dis­sem­i­na­tion Stan­dards Bul­letin Board, and SCI is from the World Bank Sta­tis­ti­cal Ca­pa­bil­ity In­di­ca­tor Data­base.

(7) As­sump­tion of ini­tial con­di­tions ( , and ). , and re­spec­tively de­note IMF fore­cast er­rors for the real eco­nomic growth rates of the U.S., Ger­many and China dur­ing pe­riod t . Con­sid­er­ing the in­flu­ence of these three coun­tries on the world econ­omy, the qual­ity of the IMF’s fore­casts on them will di­rectly in­flu­ence fore­casts on other coun­tries. Data are from the IMF World Eco­nomic Out­look Data­base.

4.2 Anal­y­sis of Em­pir­i­cal Re­sult

4.2.1 Over­all sam­ple es­ti­ma­tion re­sult

We em­ploy panel data anal­y­sis method to es­ti­mate equa­tion (4), and de­cide whether to adopt fixed-ef­fect model or ran­dom-ef­fect model based on ro­bust Haus­man test re­sult. We test two dif­fer­ent es­ti­ma­tion forms, i.e. vari­ables’ ef­fects on IMF fore­cast er­rors un­der and cir­cum­stances re­spec­tively. Aside from over­all sam­ple es­ti­ma­tion, we also con­duct sam­ple-spe­cific es­ti­ma­tion for ad­vanced and de­vel­op­ing economies to eval­u­ate the IMF’s dif­fer­ent fore­cast­ing be­hav­iors for them. In over­all sam­ple and sam­ple-spe­cific es­ti­ma­tions, we first con­sider the im­pact of SDDS adop­tion on IMF fore­cast er­rors, and then in­tro­duce the World Bank SCI to ver­ify data avail­abil­ity’s im­pact on fore­casts. Over­all sam­ple es­ti­ma­tion re­sult is shown in Ta­ble 2.

Es­ti­ma­tion co­ef­fi­cient of is sig­nif­i­cantly pos­i­tive, which shows that the IMF tends to give op­ti­mistic fore­casts on mem­ber coun­tries that re­ceive IMF loans. The larger amount of IMF loans re­ceived by a mem­ber coun­try, the more likely it is to re­ceive an op­ti­mistic fore­cast.

Es­ti­ma­tion co­ef­fi­cient of is pos­i­tive, and passes sig­nif­i­cance test on most oc­ca­sions, which means that coun­tries with greater GDP are more likely to re­ceive op­ti­mistic fore­casts from the IMF. These re­sults are con­sis­tent with ex­pec­ta­tions. The IMF tends to is­sue op­ti­mistic fore­casts on coun­tries that rep­re­sent a sig­nif­i­cant share of the world econ­omy, so as to guide world eco­nomic ex­pec­ta­tions and pro­mote world eco­nomic de­vel­op­ment.

For po­lit­i­cal re­la­tions, co­ef­fi­cients of both and are nega­tive and in­signif­i­cant, which im­plies that mem­ber coun­tries’ di­rect po­lit­i­cal re­la­tions with the IMF have an in­signif­i­cant ef­fect on the IMF fore­casts. How­ever, the es­ti­ma­tion co­ef­fi­cient of mem­ber coun­tries sim­i­lar­ity’ with the U.S. votes at the UN General As­sem­bly ( ) is sig­nif­i­cantly pos­i­tive, which means that the closer a coun­try is to the po­lit­i­cal stance of the United States, the eas­ier it be­comes to re­ceive an op­ti­mistic fore­cast from the IMF. Why do mem­ber coun­tries’ di­rect po­lit­i­cal re­la­tions with the IMF have an in­signif­i­cant ef­fect on IMF fore­cast er­rors and their in­di­rect po­lit­i­cal re­la­tions with the IMF have a rather sig­nif­i­cant ef­fect? We be­lieve that only the IMF’s ma­jor share­hold­ers are able to in­flu­ence the IMF’s fore­cast be­hav­iors. For in­stance, Stein­wand and Stone (2008) finds through lit­er­a­ture sur­vey that in var­i­ous stages of the IMF’s lend­ing ac­tiv­i­ties, only the IMF’s ma­jor share­hold­ers may ex­ert a sig­nif­i­cant in­flu­ence. That is to say, most mem­ber coun­tries are not able to in­flu­ence IMF de­ci­sions based on their own sta­tus. In­stead, they fight for fa­vor­able de­ci­sions by co­or­di­nat­ing po­lit­i­cal stances with the IMF’s ma­jor share­hold­ers, es­pe­cially the United States2.

Es­ti­ma­tion co­ef­fi­cient of with re­spect to data avail­abil­ity is nega­tive, which sug­gests that a coun­try’s adop­tion of the IMF’s SDDS helps re­duce our fore­cast er­rors and in­crease IMF fore­cast ac­cu­racy. Join­ing the IMF’s SDDS means that a coun­try passes the “good sta­tis­ti­cal prac­tice” test, and

un­der­takes to com­ply with good prac­tices re­gard­ing data cov­er­age, ap­pli­ca­tion fre­quency and time­li­ness,

3 chan­nels for pub­lic ac­cess to data, data au­then­tic­ity and data qual­ity. This pro­vides the IMF with timely and ac­cu­rate data to make fore­casts.

Er­rors of IMF fore­casts for ma­jor economies will also af­fect fore­casts on other coun­tries. Es­ti­ma­tion co­ef­fi­cients of IMF fore­cast er­rors for the U.S., Ger­many and China are all pos­i­tive, which shows that in case of er­rors of IMF fore­casts on these coun­tries, fore­cast er­rors for other coun­tries will in­crease. This in­di­cates that fore­casts on ma­jor economies play a piv­otal role in IMF fore­casts. Co­ef­fi­cients of fore­cast er­rors for China and Ger­many are sig­nif­i­cant, which means that the fore­cast ac­cu­ra­cies of these two coun­tries are par­tic­u­larly im­por­tant.

4.2.2 Sam­ple-spe­cific es­ti­ma­tion re­sult

To un­cover the IMF’s dif­fer­ent in­cli­na­tions in fore­cast­ing ad­vanced and de­vel­op­ing economies, we carry out sep­a­rate es­ti­ma­tions for such economies (Ta­ble 3). Ad­vanced and de­vel­op­ing economies are clas­si­fied ac­cord­ing to the IMF’s World Eco­nomic Out­look Data­base.

Over­all, the sam­ple-spe­cific es­ti­ma­tion re­sult is sim­i­lar to the over­all re­sult, but still of­fers some dif­fer­ent dis­cov­er­ies.

First, it is eas­ier for coun­tries that re­ceive IMF loans to be fore­casted op­ti­misti­cally, which is con­sis­tent with the es­ti­ma­tion re­sults of the over­all sam­ples. How­ever, the es­ti­ma­tion co­ef­fi­cient of ad­vanced economies is sig­nif­i­cantly higher and more sig­nif­i­cant than that of de­vel­op­ing economies. This im­plies that, com­pared with de­vel­op­ing economies, ad­vanced economies that re­ceive IMF loans are more likely to be fore­casted op­ti­misti­cally by the IMF. The IMF’s fore­cast bias can be at­trib­ut­able to var­i­ous rea­sons. Com­pared with de­vel­op­ing economies, ad­vanced economies gen­er­ally have greater voice and vot­ing rights in the IMF’s gover­nance struc­ture, and such rights are more con­cen­trated and eas­ier to in­flu­ence the IMF’s fore­cast ac­tiv­i­ties. Also, com­pared with de­vel­op­ing economies, the IMF more ur­gently needs to jus­tify loan pro­grams to ad­vanced economies through over­es­ti­mate. The rea­son is that loan pro­grams to ad­vanced economies gen­er­ally re­quire larger amounts of cap­i­tal.

Se­cond, the co­ef­fi­cient of is pos­i­tive, which is sim­i­lar to the es­ti­ma­tion re­sult of to­tal sam­ples. The es­ti­ma­tion co­ef­fi­cient for de­vel­op­ing economies is rel­a­tively sig­nif­i­cant, but co­ef­fi­cient for ad­vanced economies is in­signif­i­cant.

Third, with re­spect to mem­ber coun­tries’ di­rect po­lit­i­cal re­la­tions with the IMF, the group-spe­cific es­ti­ma­tion re­sult is sim­i­lar to the over­all es­ti­ma­tion re­sult. For ad­vanced and de­vel­op­ing economies, the co­ef­fi­cients of both and are nega­tive and in­signif­i­cant, which shows that mem­ber coun­tries’ di­rect po­lit­i­cal re­la­tions with the IMF do not have any sig­nif­i­cant in­flu­ence on IMF fore­casts.

Fourth, for mem­ber coun­tries’ in­di­rect po­lit­i­cal re­la­tions with the IMF, the es­ti­ma­tion co­ef­fi­cient of sim­i­lar­ity be­tween de­vel­op­ing economies’ votes and U.S. votes at the UN General As­sem­bly ( ) is sig­nif­i­cantly pos­i­tive, which is con­sis­tent with the es­ti­ma­tion re­sult of the over­all sam­ples. The im­pli­ca­tion is that the closer de­vel­op­ing economies are to the po­lit­i­cal stances of the United States, the eas­ier it is for them to be fore­casted op­ti­misti­cally by the IMF. How­ever, the es­ti­ma­tion co­ef­fi­cient of ad­vanced economies is sig­nif­i­cantly nega­tive, that is, the sim­i­lar­ity of po­lit­i­cal stances be­tween other ad­vanced economies and the United States will not cause the IMF to over­es­ti­mate them. The dif­fer­ence in the es­ti­ma­tion co­ef­fi­cients sug­gest that in the IMF’s fore­cast ac­tiv­i­ties, the United States as a ma­jor share­holder at­taches greater im­por­tance to whether de­vel­op­ing economies back the U.S. po­lit­i­cal stances.

Fifth, with re­spect to data avail­abil­ity, dif­fer­ences also ex­ist in the es­ti­ma­tion re­sults. The es­ti­ma­tion co­ef­fi­cient of ad­vanced economies’ re­mains nega­tive, which is con­sis­tent with the es­ti­ma­tion re­sult of the over­all sam­ples, but both es­ti­ma­tion co­ef­fi­cients are in­signif­i­cant. How­ever, the es­ti­ma­tion co­ef­fi­cient for de­vel­op­ing economies is pos­i­tive but in­signif­i­cant as well. This may have to do with the group­ing of sam­ples. For in­stance, among the 129 de­vel­op­ing economies as sam­ples, the num­ber of coun­tries adopt­ing the IMF’s SDDS in­creased from 22 to 36 dur­ing the ob­ser­va­tion pe­riod, ac­count­ing for 17.1% to 27.9% re­spec­tively of the to­tal sam­ples. Among the 31 ad­vanced economies as sam­ples, the num­ber of coun­tries adopt­ing the IMF’s SDDS in­creased from 27 to 30, ac­count­ing for 87.1% to 96.8% re­spec­tively. Such dis­e­qui­lib­rium in data dis­tri­bu­tion led to dif­fer­ences in es­ti­ma­tion re­sults. To eval­u­ate the data avail­abil­ity’s ef­fects on fore­cast er­rors, this paper sub­se­quently con­ducted es­ti­mates based on the World Bank’s Sta­tis­ti­cal Ca­pa­bil­ity In­di­ca­tor (SCI).

Sixth, the es­ti­ma­tion co­ef­fi­cients of IMF fore­cast er­rors for the United States, Ger­many, and China are all pos­i­tive, which is con­sis­tent with the es­ti­ma­tion re­sult of the over­all sam­ples. From

the per­spec­tive of es­ti­ma­tion co­ef­fi­cient sig­nif­i­cance, the co­ef­fi­cients of fore­cast er­rors for Ger­many and China with re­spect to ad­vanced economies are both rel­a­tively sig­nif­i­cant. The im­pli­ca­tion is that fore­casts on ad­vanced economies are more in­flu­enced by fore­cast er­rors for Ger­many and China. With re­spect to de­vel­op­ing economies, only the co­ef­fi­cient of fore­cast er­rors for Ger­many is rel­a­tively sig­nif­i­cant, which means that fore­cast er­rors for the Ger­man econ­omy have the great­est in­flu­ence on fore­casts on de­vel­op­ing economies.

4.2.3 Retest of data avail­abil­ity’s ef­fects on fore­cast er­rors

To fur­ther test data avail­abil­ity’s ef­fects on fore­cast er­rors, we sub­sti­tute whether a coun­try adopts the IMF’s SDDS ( ) with the World Bank’s Sta­tis­ti­cal Ca­pa­bil­ity In­di­ca­tor for var­i­ous coun­tries (

) to re-es­ti­mate equa­tion (4). The World Bank’s SCI is a com­pre­hen­sive score of a coun­try’s sta­tis­ti­cal sys­tem based on 25 in­di­ca­tors and the three di­men­sions of sta­tis­ti­cal method­ol­ogy, data sources and pe­ri­od­ic­ity

and time­li­ness. Each year, the sta­tis­ti­cal ca­pa­bil­i­ties of over 140 de­vel­op­ing coun­tries are eval­u­ated4. Com­pared with , has two sta­tis­ti­cal char­ac­ter­is­tics: First, the value range is 0-100, which is un­like the dummy vari­able of whose value is ei­ther 0 or 1. Se­cond, a coun­try’s sta­tis­ti­cal ca­pa­bil­ity de­picted by will in­crease or de­crease with the pas­sage of time. If we fol­low the stan­dard of whether a coun­try adopts SDDS, un­less un­der spe­cial cir­cum­stances, a coun­try nor­mally will not exit the stan­dard, i.e. will only de­pict the process from 0 to 1 and with­out any change from 1 to 0. There­fore, sub­sti­tut­ing with for es­ti­ma­tion helps fur­ther ver­ify the ro­bust­ness of es­ti­ma­tion. Based on data avail­abil­ity, there are 120 se­lected sam­ples es­ti­mated by , all of which are de­vel­op­ing coun­tries dur­ing the pe­riod from 2005 to 2015.

Ta­ble 4 shows the es­ti­ma­tion re­sult after the sub­sti­tu­tion of by . Es­ti­ma­tion co­ef­fi­cient of is sig­nif­i­cantly nega­tive, which sug­gests that a coun­try’s im­prove­ment of sta­tis­ti­cal ca­pa­bil­ity helps re­duce IMF fore­cast er­rors and in­crease IMF fore­cast ac­cu­racy. This con­firms this paper’s prior hy­poth­e­sis. But co­ef­fi­cient of IMF loan pro­grams is nega­tive and in­signif­i­cant. The im­pli­ca­tion is that loan pro­grams have an in­signif­i­cant ef­fect on the IMF’s eco­nomic fore­casts on de­vel­op­ing economies. This is con­sis­tent with sam­ple-spe­cific es­ti­ma­tion re­sult. Es­ti­ma­tion re­sults of other vari­ables are also gen­er­ally con­sis­tent with our prior es­ti­mates. If a coun­try is closer to the U.S. po­lit­i­cal stance at the UN

General As­sem­bly, it is more likely to be fore­casted op­ti­misti­cally by the IMF. IMF es­ti­ma­tion er­rors for ma­jor economies still have a pos­i­tive in­flu­ence on es­ti­ma­tion er­rors for other economies. This ver­i­fies the ro­bust­ness of this paper’s es­ti­mates.

5. Con­clud­ing Re­marks

This paper cre­ates a po­lit­i­cal econ­omy frame­work for an­a­lyz­ing the IMF’s fore­cast er­rors from three as­pects (fore­cast method and in­for­ma­tion, and po­lit­i­cal fac­tor), and tests the ef­fects of dif­fer­ent fac­tors on IMF fore­casts through an em­pir­i­cal anal­y­sis of 160 IMF mem­ber coun­tries dur­ing 2003-2015 with the ex­am­ple of real eco­nomic growth in­di­ca­tor. This paper’s key con­clu­sions are as fol­lows.

First, it is eas­ier for IMF mem­ber coun­tries re­ceiv­ing IMF loans to be fore­casted op­ti­misti­cally by the IMF. Sam­ple-spe­cific es­ti­ma­tion re­veals that, com­pared with de­vel­op­ing economies, ad­vanced economies re­ceiv­ing IMF loans will be fore­casted more op­ti­misti­cally by the IMF.

Se­cond, po­lit­i­cal fac­tor has a sig­nif­i­cant ef­fect on IMF fore­cast er­rors. The closer a coun­try is to the U.S. votes at the UN General As­sem­bly, the more likely it is for the coun­try to be fore­casted op­ti­misti­cally by the IMF. How­ever, IMF mem­ber coun­tries’ di­rect po­lit­i­cal re­la­tions with the IMF (quo­tas at the IMF and seats at the Ex­ec­u­tive Board) has an in­signif­i­cant ef­fect on IMF fore­cast er­rors.

Third, in­creas­ing IMF mem­ber coun­tries’ data avail­abil­ity helps im­prove IMF fore­casts. Judg­ing by over­all sam­ple es­ti­ma­tion, a mem­ber coun­try’s adop­tion of the IMF’s SDDS is con­ducive to re­duc­ing IMF fore­cast er­rors. In sam­ple-spe­cific es­ti­mates, how­ever, such a re­la­tion­ship is in­signif­i­cant. Thus, this paper con­ducts a retest based on the World Bank’s Sta­tis­ti­cal Ca­pa­bil­ity In­di­ca­tor (SCI). Re­sult shows that the im­prove­ment of sta­tis­ti­cal ca­pa­bil­ity helps re­duce fore­cast er­rors, which fur­ther ver­i­fies the role of data avail­abil­ity in the IMF’s fore­casts.

Fourth, as­sump­tions of other con­di­tions will also af­fect the IMF’s fore­casts. IMF fore­cast er­rors for ma­jor coun­tries (in this case, the United States, Ger­many and China) will mag­nify fore­cast er­rors for other coun­tries. Nev­er­the­less, the ef­fect varies across dif­fer­ent coun­tries. For ad­vanced economies, the co­ef­fi­cients of fore­cast er­rors for Ger­many and China are rel­a­tively sig­nif­i­cant. For de­vel­op­ing economies, the co­ef­fi­cient is rel­a­tively sig­nif­i­cant only for eco­nomic fore­casts on the Ger­man econ­omy. The im­pli­ca­tion is that im­prov­ing the fore­cast ac­cu­racy of these two coun­tries is par­tic­u­larly im­por­tant.

Based on this paper’s con­clu­sions, we pro­pose the fol­low­ing rec­om­men­da­tions to im­prove the IMF’s fore­cast qual­ity. First, the IMF should cre­ate a more in­de­pen­dent fore­cast pro­ce­dure to avoid in­ter­fer­ence of in­ter­nal and ex­ter­nal fac­tors. To make its fore­casts more re­li­able, the IMF should re­duce the im­pacts of po­lit­i­cal fac­tors and or­ga­ni­za­tional pref­er­ences on its eco­nomic fore­casts. Specif­i­cally, the IMF might in­tro­duce ex­ter­nal mech­a­nisms to en­hance fore­cast su­per­vi­sion and re­search, and re­strain its in­ter­nal be­hav­iors through such ex­ter­nal mech­a­nisms.

Se­cond, the IMF should con­tinue im­prov­ing the fore­cast data qual­ity. IMF mem­ber coun­tries should be helped to stan­dard­ize data re­port­ing and im­prove sta­tis­ti­cal ca­pa­bil­i­ties and data re­port­ing qual­ity through bi­lat­eral sur­veil­lance and tech­ni­cal sup­port.

Third, the IMF should im­prove its fore­cast ac­cu­racy for ma­jor economies by staffing and train­ing coun­try re­searchers for ma­jor economies and en­hanc­ing coun­try of­fices’ fore­cast ca­pa­bil­i­ties. The IMF should also en­hance com­mu­ni­ca­tions with IMF mem­ber coun­try gov­ern­ments and pri­vate sec­tors to re­duce fore­cast er­rors for ma­jor economies.

Source: the World Eco­nomic Out­look (WEO) pub­lished by the IMF. Notes: Fore­cast er­rors are based on the spring fore­cast of the World Eco­nomic Out­look. Fig­ure 1: Er­rors in the IMF’s Fore­casts on Mem­ber Coun­tries’ Real Eco­nomic Growth (%)

Newspapers in English

Newspapers from China

© PressReader. All rights reserved.