Moral Hazard and Peer Monitoring in a Laboratory Microfinance Experiment *

Transcription

1 Moral Hazard and Peer Montorng n a Laboratory Mcrofnance Experment * Tmothy N. Cason, Lata Gangadharan and Pushkar Matra Revsed: November 2011 Abstract Ths paper reports the results from a laboratory mcrofnance experment of group lendng n the presence of moral hazard and (costly) peer montorng. We compare peer montorng treatments n whch credt s provded to members of the group to ndvdual lendng treatments wth lender montorng. We fnd that f the cost of peer montorng s lower than the cost of lender montorng, peer montorng results n hgher loan frequences, hgher montorng and hgher repayment rates compared to lender montorng. In the absence of montorng cost dfferences, however, lendng, montorng and repayment behavour s mostly smlar across group and ndvdual lendng schemes. Wthn group lendng, contrary to theoretcal predctons, smultaneous and sequental lendng rules provde equvalent emprcal performance. JEL Classfcaton: G21, C92, O2. Key words: Group Lendng, Montorng, Moral Hazard, Laboratory Experment, Credt, Development * We have benefted from comments by Dyut aneree, Shyamal Chowdhury, Roland Hodler, Jeorg Oechssler, Kunal Sengupta, Tom Wlkenng, Chkako Yamauch, partcpants at semnars and workshops at Monash Unversty, Unversty of Melbourne, the Australan Natonal Unversty, the Unversty of Adelade, Unversty of Sydney, Jadavpur Unversty, Kolkata, the Indan Statstcal Insttute, Delh, partcpants at ESA Internatonal meetngs, NEUDC Conference, the Econometrc Socety Australasan Meetngs, the Econometrc Socety World Congress and two anonymous referees for ther comments and suggestons. We would lke to thank Gautam Gupta for hs assstance n organzng the sessons at Jadavpur Unversty, Kolkata. Tana Dey, Smon Hone, Vnod Mshra and Roman Sheremeta provded excellent research assstance. The usual caveat apples. Tmothy Cason, Department of Economcs, Krannert School of Management, Purdue Unversty, West Lafayette, IN , USA. Emal: cason@purdue.edu Lata Gangadharan, Department of Economcs, Monash Unversty, Clayton Campus, VIC 3800, Australa. Emal: Lata.Gangadharan@monash.edu Pushkar Matra, Department of Economcs, Monash Unversty, Clayton Campus, VIC 3800, Australa. Emal: Pushkar.Matra@monash.edu

2 1. Introducton The last few decades have wtnessed the development of nnovatve and hghly successful mcrofnance mechansms for the provson of credt to the poor. The most common of these s group lendng. Rather than use ndvdual lendng rules where the bank (or the lender) makes a loan to an ndvdual who s solely responsble for ts repayment, n group lendng the bank makes a loan to an ndvdual who s a member of a group and the group s ontly lable for each member s loans. If the group as a whole s unable to repay the loan because some members default on ther repayment, all members of the group become nelgble for future credt. The Grameen ank n angladesh, the well known mcrofnance nsttuton (MFI) that was the poneer of such group lendng programs, reports that as of 2008, only 2.08% of ts loans are overdue (see The success of the Grameen ank has led polcy makers and Non-Government Organsatons around the world to ntroduce smlar schemes. 1 Whle n recent years, several MFIs have moved on from group based lendng programs, group lendng contnues to be the most commonly used mechansm n the context of credt provson by MFIs. Indeed the current trend towards ndvdual lendng programs makes a systematc study of the performance of lenders and borrowers n ndvdual and group lendng programs topcal and mportant from an academc and a polcy pont of vew. The am of ths paper s to examne lendng, montorng and repayment behavour n group and ndvdual lendng schemes, usng expermental methods. We report the results from a laboratory experment of group lendng n the presence of moral hazard and (costly) peer montorng. 2 We fnd 1 As of 2007, mcrofnance nsttutons were servng around 150 mllon people around the world (Gne et al. (2010)). The 2006 Nobel Prze for Peace to mcrofnance poneer Muhammed Yunus also put the success of mcrofnance n the world spotlght was desgnated by the Unted Natons as the year of mcrofnance. Whle mcrofnance programs are most wdespread n less developed countres they are by no means confned to them. These programs have been ntroduced n transton economes such as osna and Russa and n developed countres such as Australa, Canada and the US (see Conln (1999), Armendarz de Aghon and Morduch (2000), Armendarz de Aghon and Morduch (2005) and Fry et al. (2008)). Mcro-lendng s ncreasngly movng from non-proft towards a proft-makng enterprse, wth large banks such as Ctgroup now backng such loans (ellman (2006)). 2 We focus on nformatonal asymmetres due to moral hazard rather than those due to adverse selecton. In partcular we restrct attenton to exogenously formed groups and leave the ssue of endogenous group formaton (postve assortatve matchng) for future research. Ghatak (2000), Van Tassel (1999), and Armendarz de Aghon and Goller (2000) dscuss theoretcal models on how group lendng solves the problem of adverse selecton. 1

3 that smultaneous and sequental credt provson to group members leads to smlar results. Compared to ndvdual lendng, however, group lendng leads to greater loan frequences, and hgher montorng and mproved repayment rates f peer montorng s less costly than lender montorng. Ths mportance of montorng costs on credt market performance n our experment s consstent wth perceved advantages of group lendng n practce. The success of group lendng programs arses, n part, because they can better address the enforcement and nformatonal problems that generally plague formal sector credt n developng countres. 3 Group lendng programs typcally help solve the enforcement problem through peer montorng. Stgltz (1990) and Varan (1990) argue that snce group members are lkely to have better nformaton compared to an outsder lke, the bank, peer montorng s relatvely cheaper compared to bank montorng, leadng to greater montorng and hence greater repayment. aneree, esley and Gunnane (1994) argue that peer montorng s better at explanng the success of group lendng programs than alternatve explanatons. 4 Most emprcal studes on the determnants of repayment use data from nsttutons wth smlar lendng rules, hence there s relatvely lttle varaton to estmate the effcacy of a partcular mechansm. Thus, lackng well desgned experments, they are forced to rely on varaton n the economc envronment to dentfy the parameter of nterest, and often tmes they employ nstruments that are hard to ustfy. Also, the varaton that does exst n the feld s endogenous, whch makes t dffcult to unambguously determne causalty (Morduch (1999), Armendarz de Aghon and Morduch (2005)). Mcrofnance loans present a complex economc envronment, and the lterature does not yet provde a unfed approach to analyse contracts and borrower and lender behavor. Experments, grounded n careful theory, have an mportant role to play n ths respect. Few laboratory experments examne the mpact of specfc desgn features on the performance 3 Armendarz de Aghon and Morduch (2000), Armendarz de Aghon and Morduch (2005), Chowdhury (2005), Che (2002), Ra and Sostrom (2004) and hole and Ogden (2010) dscuss dfferent aspects of mcrofnance programs. 4 Peer montorng and peer enforcement have been observed to deter free rdng n several experments relatng to other socal dlemma stuatons, such as common pool resource envronments and the voluntary provson of publc goods. See Fehr and Gaechter (2000), arr (2001), Masclet et al. (2003), Walker and Halloran (2004), and Carpenter, owles and Gnts (2006) for expermental evdence. 2

4 of mcrofnance models. Abbnk, Irlenbusch and Renner (2006) and Seddk and Ayed (2005) examne the role of group selecton n the context of group lendng. oth experments are desgned as nvestment games where each group member nvests n a rsky proect whose outcome s known only to the ndvdual, and both fnd that self-selected groups have a greater wllngness to contrbute. Nether of these papers analyse the role of peer montorng. Ths paper contrbutes to the recent debate on ont versus ndvdual lablty n mcrofnance. In recent years, programs lke Grameen II and anco Sol have adopted ndvdual lablty, where each loan recpent s ndvdually responsble for repayment of her loans. Gne and Karlan (2009) use data from a controlled feld experment n Phlppnes and argue that convertng exstng borrowers to ndvdual lablty does not affect repayment rates; and even when groups are ntally formed under ndvdual lablty, repayment rates are no lower. The role of a group n such a scenaro s smply to act as a medator. On the other hand, programs lke Self Help groups n Inda contnue to rely on ont lablty where members are ontly responsble for loan repayment. As many mcro lenders swtch or consder swtchng from group to ndvdual-lablty loans, t s mportant to understand the mechansms determnng outcomes n the dfferent scenaros. Achevng ths goal s partcularly challengng as t can be affected by nformaton, montorng, and proect choce. 5 We fnd that n the envronment that we consder, the results depend on the cost of montorng. If the cost of montorng under an ndvdual lablty program s no dfferent to that under a ont lablty program, then the two provde almost equvalent performance. If however, the cost under peer montorng s lower, compared 5 Feld experments whle feasble are dffcult to mplement and sometmes come at the cost of some loss of expermental control. For example, spll overs from one vllage to another or from the treatment group to the control group could create nose n the data. Snce groups are self-formed n the feld the benefts of peer montorng could also be over-estmated due to assortatve matchng. It mght therefore be dffcult to separate out the effects of peer montorng and group selecton usng feld data. Ths s not a problem n our laboratory experment, whch features strctly random assgnment. That sad, n recent years there have been a number of nnovatve feld experments dealng wth dfferent aspects of mcrofnance. Gne et al. (2010), Fscher (2008), Kono (2006), Cassar, Crowley and Wydck (2007) report artefactual feld experments whch place non-student subects n stylzed mcrofnance envronments smlar to controlled laboratory studes. Feld and Pande (2008), aneree et al. (2009), Gne and Karlan (2009), Karlan and Znman (2010), Fegenberg, Feld and Pande (2011) on the other hand descrbe randomzed nterventons for actual mcrofnance programs. Our paper complements ths rapdly growng lterature by drectly testng some of the theoretcal predctons relatng to ndvdual and group lendng and the cost of montorng, whch s dffcult to do n the feld. 3

5 to the cost of lender montorng, then ont lablty domnates. We also document dfferences n group and ndvdual lendng treatments that could be attrbuted to group-based responsblty and trust. 2. Theoretcal Framework Overvew Consder a scenaro where two borrowers requre one unt of captal (say $1) each for nvestng n a partcular proect. The bank, whch provdes ths captal n the form of a loan, can ether make the loan to an ndvdual (ndvdual lendng) or t can loan to the borrowers as a group (group lendng). In the case of group lendng the borrowers are ontly responsble for the repayment of the loan. oth the borrowers and the lenders are assumed to be rsk-neutral and am to maxmse ther profts. orrowers can nvest n two dfferent types of proects: one proect has a large verfable ncome and no nonverfable prvate beneft (proect S), whle the other has a large non-verfable prvate beneft and no verfable ncome (proect R). The bank prefers the frst proect, where t can recoup ts nvestment, but the borrowers prefer the second one. In the absence of montorng, the borrowers wll choose to nvest n the second proect and the bank, knowng ths, wll choose not to make the loan. Let us brefly descrbe the theoretcal framework, whch forms the bass of our expermental desgn and hypotheses. The framework closely follows Chowdhury (2005) and Ghatak and Gunnane (1999). Suppose that there are two borrowers: 1 and 2. If Proect S s chosen, the return s H (verfable by montorng) and f proect R s chosen, then the return s b (not verfable) wth b 4 < H. The $1 cost of each proect s fnanced by a loan from the bank (or a lender) snce the borrowers do not have any funds of ther own. When the two borrowers ( 1 and 2 ) borrow together as a group, each borrower receves $1 from the lender. The amount to be repad s r ( > 1) n the case of ndvdual lendng or 2r n the case of group lendng. We assume that ths nterest rate r s fxed exogenously. In the case of ndvdual lendng, f the borrower chooses proect S the return to the bank s r; otherwse t s 0. The return to the borrower s H r f the borrower chooses proect S, and s b f the

6 borrower chooses proect R. We assume that H r < b so that borrowers prefer proect R. anks on the other hand prefer proect S. In the case of group lendng, f both borrowers choose proect S, the return to each borrower s H r and the return to the bank s 2. r If both borrowers choose proect R, the return to each borrower s b and the return to the bank s 0. Fnally f one borrower chooses proect R and the other chooses proect S, then due to ont lablty the return to the borrower choosng proect S s 0 whle that of the borrower choosng proect R s b and the return to the bank s H. We assume that H 2r. In the case of group lendng therefore the bank s better off f both borrowers choose proect S, and each borrower has an ncentve to ensure that the other borrower chooses proect S so that the bank has an ncentve to lend. An nformatonal asymmetry arses because each borrower knows the type of hs own proect, but the lender or the other borrower n the group can fnd out the borrower s proect choce only wth costly montorng. The montorng process works as follows: orrower can, by spendng an amount c( m ) n montorng costs, obtan nformaton about the proect chosen by the other borrower n hs group ( ) wth probablty [ 0,1] m. Ths nformaton can be used by to ensure that chooses proect S. One could thnk of dfferent ways n whch montorng works n practce: nformaton acqured by the borrowers about each other s proect choce may be passed on to the lender who then uses ths nformaton to force the borrowers to choose proect S. Alternatvely, through montorng the borrowers can use some form of socal sanctons or peer punshment to ensure that the other borrower chooses proect S. The bank (lender) can also acqure ths same nformaton by spendng an amount λ c( m). We assume that λ 1 n order to capture the noton that peer montorng s less expensve than montorng by the bank. We assume for smplcty a quadratc montorng cost functon ( ) and ths quadratc functon s mplemented n the experment. 2 m c m =, 2 5

7 In practce peer montorng s usually less costly than drect lender montorng; ndeed, ths cost advantage s regarded as one of the man benefts of peer montorng. Hermes and Lensnk (2007) argue that the hgher observed repayment rates n group lendng wth peer montorng compared to ndvdual lendng wth lender montorng s drven by the greater effectveness of screenng, montorng and enforcement wthn the group. Ths could be due to the closer geographcal proxmty and close socal tes between the group members, whch translate to lower montorng costs n the case of group lendng wth peer montorng compared to ndvdual lendng wth lender montorng. Our expermental desgn also compares credt market performance when drect lender montorng and peer montorng nvolve the same montorng cost ( λ = 1). Ths allows us to examne the relatve effectveness of group lendng wth peer montorng and ndvdual lendng wth lender montorng, holdng montorng costs constant. Indvdual Lendng Frst consder ndvdual lendng (wth bank montorng). There are three stages to the game. Stage 1: ank chooses whether or not to lend $1 to the borrower. If the bank chooses not to lend, then the $1 can be put nto alternatve use, whch yelds r < 1. Stage 2: ank chooses the level of montorng, condtonal on decdng to lend. Stage 3: orrower chooses ether proect R or proect S. It s straghtforward to solve for the sub-game perfect Nash equlbrum of the game by λm backward nducton. If the bank lends, t chooses m to maxmse mr 2 2 1, whch gves m * = r. λ Therefore the expected return to the bank s 2 r 1, so the bank wll provde the loan f and only f 2λ r r λ > ;.e., f 2 r 2λ ( r 1) > +. Ths gves rse to the frst proposton: 6

8 Proposton 1: If the costs of montorng relatve to the return are suffcently low,.e., 2 r λ <, ( r ) then ndvdual lendng s feasble, and the effcent (montorng/lendng) equlbrum exsts; for montorng costs above ths threshold the unque equlbrum has no lendng. We consder two specfcatons for the montorng cost structure n the experment. In the frst the 2 r ndvdual lendng hgh cost treatment we set λ >. In the second, the ndvdual lendng low 2 r r cost treatment we set λ < 2( r + 1). Group Lendng: Smultaneous The sequence of events n group lendng s as follows: ( ) Stage 1: ank chooses whether or not to lend $2 to the group. There s ont lablty, so that f one borrower fals to meet hs oblgatons, then f the other borrower has verfable ncome he must pay back the bank for both borrowers. If the bank chooses not to lend, then the $2 can be put nto alternatve use, whch yelds r < 1 per dollar. Stage 2: The borrowers smultaneously choose the level of peer montorng, Stage 3: oth borrowers choose ether proect R or proect S. Note that here both montorng and lendng s smultaneous. Agan the sub-game perfect Nash m. equlbrum s solved by backward nducton. orrower wll choose montorng m to maxmze m m m( H r) ( 1 m) b ( 1 m) m 0 ( 1 m) b The frst order condton s: m ( H r) m 0 ( ) 0 =. Lkewse the frst order condton for borrower 2 s: m H r m =. Clearly m * = m * = 0 s a Nash equlbrum. We call ths the neffcent (zeromontorng/zero-lendng) equlbrum. In ths case there s a strategc complementarty between the montorng levels of the two borrowers. A borrower knows that f the other borrower montors and he does not, then he wll end up wth a payoff of 0. If however the other borrower does not montor then 7

9 he has no ncentve to montor as well. Hence ont lablty and peer montorng would not solve the moral hazard problem. Remember however that [ 0,1] m. Now consder 's reacton functon m = m ( H r). If H r > 1, there exsts a m = ˆm < 1 such that the best response s m = 1 for m > ˆm. So 's complete reacton functon can be wrtten as: m = m ( H r), f m ˆm 1, f m > ˆm In ths case the corner soluton m = m = 1 s also a Nash equlbrum (and the dervatve of the ** ** borrowers value functon s strctly postve). We can call ths the effcent (montorng/lendng) equlbrum. Fgure 1 presents the reacton functons for H r = It s mportant to note that the reacton functons are upward slopng. We wll return to ths ssue n the emprcal results. The lender wll choose to lend f her expected payoff from lendng exceeds that from not lendng. The lender wll therefore choose to lend f (see Appendx A for a dervaton of ths condton): ( ) 2 ** ** m r H + m H > 1+ r The bank s payoffs n these two montorng game equlbra determne whether t wll lend. For the neffcent (0,0) case, the expected payoff to the bank s 2< 2r and group lendng s not feasble. The payoff to both borrowers n ths case s 0. On the other hand, for the effcent ( 1,1 ) case, the payoff to the bank s 2r 2> 2r and the payoff to both borrowers s H r 1. Clearly 2 m = m = 1 ** ** s the payoff-domnant equlbrum. Although ths also makes t a focal pont equlbrum (Schellng (1980), p. 291), prevous expermental evdence ndcates that ths s not a suffcent condton for behavoural equlbrum selecton (e.g., Van Huyck, attalo and el (1990)). Proposton 2: If H r > 1 and agents coordnate on the payoff-domnant Nash equlbrum, then under a smultaneous group lendng scheme lenders choose to make loans, borrowers choose a hgh level of 8

10 montorng and repayment rates are hgh leadng to an effcent (montorng/lendng) equlbrum. However, an neffcent zero-montorng equlbrum wth no lendng also exsts. Group Lendng: Sequental An alternatve to smultaneous lendng s to lend sequentally to group members wth the order chosen randomly. Here ntally only one (randomly chosen) member of the group receves a loan. Dependng on whether ths loan s repad, the bank decdes whether or not to lend to the other member of the group. Ths ncorporates dynamc ncentves, whch have become ncreasngly popular among researchers and practtoners n mcrofnance. 6 The sequence of events s as follows: Stage 1: ank chooses whether or not to lend $1 to one of the members of the group. The other dollar can be put nto alternatve use, whch yelds r < 1 f the actual proect choce of the frst randomly chosen borrower s proect R and the second borrower does not receve the loan. Stage 2: The borrowers smultaneously choose ther levels of montorng Stage 3: One of the borrowers s chosen at random (wth probablty α ) to receve the frst loan, f the bank lends. Ths borrower, earns b and nether, decdes whether to nvest n R or S. If m. nvests n proect R, then he (the second borrower) nor the bank receves anythng. The game stops here. Note that f the bank chooses not to lend to ether borrower, then the $2 can be put nto alternatve use, whch yelds r < 1 per dollar. Stage 4: The game moves to round 2 only f bank lends $1 to montorng, then (the frst borrower) nvests n proect S n round 1. The who nvests n ether proect R or proect S (of course f has to nvest n proect S). was successful n her 6 Dynamc ncentves mean that banks make future loan accessblty contngent on full repayment of the current loan to prevent strategc default. Ray (1998) argues that ths knd of sequental lendng mnmzes the contagon effect assocated wth ndvdual default. Sequental lendng can also mnmze the potental of coordnaton falure. Chowdhury (2005) and Anket (2006) argue that n a smultaneous group lendng scheme wth ont lablty and costly montorng, peer montorng by borrowers alone s nsuffcent and that sequental lendng that ncorporates dynamc ncentves s essental to mprove repayment rates. 9

11 If (the frst borrower) nvests n proect S n round 1, we assume that the bank collects the entre output H and holds on to t. If (the second borrower) also nvests n proect S, the bank collects r from and returns H r to. The earnngs of each borrower then are H r and the bank s earnngs are 2( r 1). If nvests n proect R, the bank collects 0 from entre output of f, whch s H. So earns 0, nvests n proect R n round 1, then and retans the earns b, and the bank s earnngs are H 2. Fnally does not receve a loan (the bank puts the second dollar to alternatve use). Earnngs of are b, earnngs of are 0, and the bank s earnngs are 1+ r. Ths happens rrespectve of the proect chosen by. The reacton functons for the two borrowers are symmetrc and are gven by (see Appendx for a dervaton of the reacton functons): Solvng out and smplfyng we get ( 1 α ) ( 1 α ) ( 1 α ) ( 1 α ) m = m H r b + b m = m H r b + b ( 1 α ) b ( α ) m = = m = m 1 H r 1 b Thus a unque and postve level of montorng exsts as long as H r 1 α < 1 b,7 although an nteror soluton s not defned f 1+ H r ( 1 α ) b = 0 or ( α ) 1 H r 1 b = 0. Ths postve level of montorng occurs because even f borrower does not montor, has an ncentve to montor. To see ths, suppose that receves the loan n round 1 (remember that the order of recevng the loan s determned randomly). If does not montor, wll nvest n proect R and 7 Ths condton, derved from the need for the denomnator mmedately above to be postve, smply requres that the borrowers are suffcently uncertan about the order n whch they would be chosen to be the frst and second borrower. 10

12 then wll receve a payoff of 0. y choosng a postve level of montorng, can ncrease the probablty that nvests n proect S. In ths case the game contnues onto the second round and gets the loan. Moreover, gven that s gong to montor, has an even greater ncentve to montor due to the strategc complementarty of montorng. So the sequental nature of the lendng scheme and the smultaneous choce of the level of montorng (before a borrower knows whether he s the frst or the second borrower) leads the effcent (montorng/lendng) equlbrum to be unque, as long as the equlbrum montorng levels are suffcent to provde postve net returns to the lender. 8 H r 1 Proposton 3: If α < 1 b and m m( 2r H) + ( H r 1) > r+ 1, then under sequental group lendng, a unque Nash equlbrum exsts n whch lenders choose to make loans, borrowers choose a hgh level of montorng and repayment rates are hgh leadng to an effcent (montorng/lendng) equlbrum. The symmetrc montorng rates n ths case are gven by ( 1 α ) b m = Mn 1, = m = m. An nteror soluton to the montorng rate s not defned 1 H r ( 1 α ) b f 1+ H r ( 1 α ) b = 0 or f 1 H r ( 1 α ) b = 0. The frst expresson n the f statement ensures that montorng s postve, and the second expresson ensures that the lender chooses to make loans (see Appendx C for a dervaton of ths condton). For the parameter values that we have chosen, H = 4; b= 2.5; r = 2.25; r = 0.75; α = 0.5 (see Table 1), we have a corner soluton: optmally each borrower would lke to choose m > 1, but recall that montorng s restrcted n the nterval [ 0,1 ]. Hence n equlbrum each borrower wll choose the maxmum permssble level of montorng whch s equal to 1 n our framework. At ths corner soluton, the dervatve of the borrowers value functon s strctly postve. The lender s payoff s 2r 2 = 2.5, whch exceeds the 2r = 1.5 payoff from not lendng. 8 In feld settngs, groups are often self-selected. In such a stuaton one could thnk of the montorng costs as screenng costs that the group members have to ncur pror to group formaton. y ncurrng ths cost, borrowers are able to ensure that other members of the group wll not make choces that are detrmental to the group. Possbly ncurrng a hgher screenng cost pror to group formaton gves borrowers greater leverage n affectng the proect choce of other members of the group. In the lab we replcate ths dea by makng borrowers ncur the montorng cost before they know the order n whch they receve the loan. 11

13 In Fgure 2 we present the best response of orrower to alternatve montorng rates chosen by orrower for the experment parameters. These reacton functons ndcate the choce of montorng rate that maxmzes a borrower s expected payoffs gven the montorng rate chosen by the other borrower. For example, f knew he was the frst borrower and beleved that would montor at level m = 0.1, hs expected payoff would be maxmzed at a montorng level of m = 0.2. Snce montorng decsons are made before each borrower knows whether he s the frst or the second borrower, and each knows that they wll be randomly chosen to be the frst or the second borrower wth probablty 0.5, the relevant lne s shown wth trangle labels. Irrespectve of whether one s the frst or the second borrower, the optmal response of each borrower s to choose a level of montorng hgher than that chosen by the other borrower. Consequently, for the experment parameters both borrowers have a strctly domnant strategy to choose the maxmum level of montorng. Thus the effcent (montorng/lendng) equlbrum s unque. The sequental nature of the lendng scheme and the smultaneous choce of the level of montorng lead each borrower to choose the maxmum permssble level of montorng, and knowng ths the lender wll choose to make the loan. 3. Expermental Desgn We desgned four treatments to examne the equlbrum predctons descrbed n Propostons 1 3. Two treatments were ndvdual lendng treatments, wth 12 subects randomly dvded nto groups of two wth each group consstng of one borrower and one lender. These two treatments dffer n the lender s montorng costs: hgher n the ndvdual lendng hgh cost treatment compared to the ndvdual lendng low cost treatment. The other two treatments were group lendng treatments, wth the 12 subects randomly dvded nto groups of three wth each group consstng of two borrowers and one lender. One s the smultaneous group lendng treatment and the other s the sequental group lendng treatment. The role of each subect (as a borrower or as a lender) was determned randomly and 12

14 remaned the same throughout all 40 perods of each sesson. At the end of every perod partcpants were randomly re-matched. Subects partcpated n one sesson only. The two proects avalable to borrowers, S and R, each cost $1, to be fnanced by a loan from the lender. In the ndvdual lendng treatments, the lender chose whether or not to nvest $1 nto ths loan. If the lender decded to make the loan she could montor the proect choce of the borrower by choosng to pay a montorng cost (C). In the group lendng treatments, the lender chose whether or not to nvest $2 nto the loan ($1 to each borrower). In ths case the lender could choose to make the loan to both borrowers or to nether. If the lender chose not to make loans, she earned $1.50 (or $0.75 n the ndvdual lendng treatment) for the perod. In the group lendng treatments, f the borrower receved the loan, he could montor the proect choce of the other borrower n the group by choosng to pay a montorng cost (C). oth borrowers could montor each other. If a borrower ncurred a cost C on montorng, there was a chance of m that the other borrower would be requred to choose proect S. Otherwse the other borrower could choose ether proect R or proect S. Montorng decsons were made smultaneously. In the sequental group lendng treatment, the borrowers were randomly determned to be the frst or the second borrower n the group to receve the loan. In ths case f the frst (randomly chosen) borrower s actual proect choce was R, then the lender s second dollar was automatcally allocated to her savngs account where she earned $0.75 for ths dollar. The theoretcal predctons and the parameter values used are summarzed n Table 1 (Panel A and Panel respectvely). These parameter values were chosen to satsfy the parameter restrctons n Propostons 1 3 and mplement a test of the theoretcal model. These parameters mply specfc earnngs of the borrowers and the lender, shown n Table 1, Panels C E. 13

15 We used the strategy method to elct decsons from the borrowers. 9 The use of ths method mples that the borrowers and lenders made decsons smultaneously and borrowers made ther decson before they knew whether or not they had receved the loan. In the case of sequental lendng, the borrowers made montorng decsons before they knew whether they were the frst or the second borrower n ther group to receve a loan. They dd, however, know whether they were the frst or the second borrower to receve the loan at the tme of makng ther proect choce. Our choce of random re-matchng of subects algns the expermental envronment wth the theoretcal model, whch does not feature reputaton formaton. In practce, ths makes the envronment more relevant for mcrofnance n urban slums, where groups are usually formed exogenously (see Karlan (2007) for an example). Socal captal and long term relatonshps between borrowers, whch may be mportant for the success of group based lendng programs n rural areas, are vrtually nonexstent n urban slums. A sgnfcant montorng cost dfferental between lender and peer montorng could stll exst n such an urban envronment, snce fellow borrowers lve n the same communty, but peer montorng costs are lkely to vary sgnfcantly between urban and rural settngs. One of the advantages of our expermental desgn s that t enables us to examne explctly the mplcatons of changng the cost dfferental between lender and peer montorng treatments. A second feature of our desgn s that the lendng decson s a choce varable. Ths allows us to examne lender behavour, whch mght be dffcult to do n the feld. We conducted a total of 29 sessons n Australa and Inda across these treatments wth 12 subects n each sesson, wth 5 sessons n Treatment 2 and 8 n the other three treatments. Twenty of the 29 sessons were conducted n Australa and the remanng n Inda. The 348 subects who partcpated n these sessons were graduate and undergraduate students at Monash Unversty and Unversty of Melbourne, Australa and Jadavpur Unversty, Kolkata, Inda. All subects were 9 The strategy method smultaneously asks all players for strateges (decsons at every nformaton set) rather than observng each player s choces only at those nformaton sets that arse n the course of a play of a game. Ths allows us to observe subects entre strateges, rather than ust the moves that occur n the game. 14

16 nexperenced n that they had not partcpated n a smlar experment. Subects earned payments n expermental dollars, whch were converted to local currency at a fxed and announced exchange rate. At the end of the sesson subects were pad the amount they had accumulated over the 40 perods and on average they earned AUD or ts purchasng power equvalent. 10 The z-tree software (Fschbacher (2007)) was used to conduct the experment. Each sesson lasted approxmately 2 hours, ncludng nstructon tme. The nstructons (ncluded for the smultaneous group lendng treatment n Appendx D) used the borrowng and lendng termnology employed n ths descrpton. 4. Hypotheses to be tested The experments were desgned to test the followng theoretcal hypotheses, whch follow from propostons 1 3: Hypothess 1 (H1) lendng: The lendng rate s a. strctly lower for the ndvdual lendng hgh cost treatment compared to the other three treatments; b. at least as hgh n the sequental group lendng treatment compared to the smultaneous group lendng treatment; and c. at least as hgh n the ndvdual lendng low cost treatment compared to the smultaneous group lendng treatment. Hypothess 2 (H2) montorng: The montorng rate a. s strctly lower for ndvdual lendng hgh cost treatment compared to the other three treatments; b. at least as hgh n the sequental group lendng treatment compared to the smultaneous group lendng treatment; c. at least as hgh n the ndvdual lendng low cost treatment compared to the smultaneous group lendng treatment. Hypothess 3 (H3) repayment: The repayment rate s a. s strctly lower for ndvdual lendng hgh cost treatment compared to the other three treatments; b. at least as hgh n the sequental group lendng treatment compared to the smultaneous group lendng treatment; c. at least as hgh n the ndvdual lendng low cost treatment compared to the smultaneous group lendng treatment. 10 At the tme of the experment, 4 Australan dollars were worth about 3 U.S. dollars. 15

17 Part a of each hypothess concerns the change n montorng cost, holdng constant the aspect of ndvdual lendng wth lender montorng; 11 Part b compares the two forms of group lendng; Part c compares the outcomes under smultaneous group lendng to ndvdual lendng wth lender montorng, holdng montorng cost constant. In summary, for all three performance measures the treatments are ordered as: ndvdual lendng low cost treatment = sequental group lendng treatment smultaneous group lendng treatment > ndvdual lendng hgh cost treatment. The weak nequalty n parts b and c of these hypotheses follow from the theoretcal predctons and parameter choces, whch mply that the effcent (montorng/lendng) equlbrum s unque n the sequental group lendng and ndvdual lendng low cost treatments, but both effcent and neffcent (zero-montorng/zero-lendng) equlbra exst n the smultaneous group lendng case. Thus, theoretcally outcomes could be less effcent n the smultaneous group lendng treatment compared to sequental group lendng treatment and the ndvdual lendng low cost treatment. The expermental results wll reveal whether ths behavoural dfference arses emprcally. 5. Results We present our results n the next three subsectons, wth each subsecton addressng a specfc aspect of the program performance: lendng, montorng, and repayment. In each case we present conservatve non-parametrc Mann-Whtney rank sum tests for treatment dfferences that requre mnmal statstcal assumptons and are based on only one ndependent summary statstc value per sesson. We also report estmates from multvarate parametrc regresson models that can dentfy the contrbuton of dfferent factors on lender and borrower behavour. Our results are summarzed n Table 6 below. Lendng Fgure 3 presents the average proporton of lenders makng loans n the dfferent perods, by treatment. 11 Strctly speakng n Hypotheses 2 and 3, Part a does not derve from an equlbrum predcton. Ths s because n equlbrum there should be no lendng n the ndvdual lendng hgh cost treatment. Snce montorng and repayment s condtonal on lendng, they are not defned n equlbrum for ths treatment. We nevertheless nclude Part a n these two hypotheses because n the experment we see postve lendng rate n the ndvdual lendng hgh cost treatment, so montorng rates and repayment rates are defned emprcally, although they should be low off the equlbrum path. 16

18 Clearly the average proporton of lenders makng loans s substantally lower at every perod for the ndvdual lendng hgh cost treatment but there s very lttle dfference n the early perods between the ndvdual lendng low cost and group lendng treatments. However the lendng rate n the last 5 perods s sgnfcantly lower n the ndvdual lendng low cost treatment compared to the group lendng treatments (Table 2, Panel A). Ths suggests that over tme lendng rates are modestly lower n ndvdual lendng compared to group lendng even holdng montorng costs constant (though the dfferences are not statstcally sgnfcant usng a non-parametrc ranksum test usng sesson level averages as the unt of observaton). Dfferences n montorng costs across the dfferent montorng regmes exacerbate the dfferences n lendng rates between ndvdual and group lendng programs, as the ndvdual lendng hgh cost treatment has by far the lowest lendng rate. Subects partcpated n the experment for 40 perods, allowng us to examne ther behavour over tme more systematcally usng panel regressons. Table 3 presents two econometrc models of the lenders loan decsons. These panel regressons ncorporate a random effects error structure, where the subect (lender) represents the random effect and the standard errors are clustered at the sesson level to account for potental sesson level unobserved heterogenety. The dependent varable s 1 f the lender chooses to lend. We present the results from two dfferent specfcatons. Specfcaton 1 ncludes a dummy for group lendng, and specfcaton 2 replaces ths wth separate dummes for the two group lendng treatments. oth specfcatons nclude a dummy for the ndvdual lendng low cost treatment, and the reference category s always the ndvdual lendng hgh cost treatment. The confguraton of sgn and sgnfcance of 1 t, 1 INDVLOWCOST t and 1 GROUP t ndcate that lendng decreased over tme n the two ndvdual lendng treatments, but ncreased over tme n the two group lendng treatments, relatve to the reference category (ndvdual lendng hgh cost treatment). Lendng rates are sgnfcantly hgher n the group lendng treatments compared to the 17

19 ndvdual lendng low cost treatment. 12 The probablty of lendng n perod t s sgnfcantly lower f the lender receved negatve earnngs n perod t 1, whch provdes some smple evdence of renforcement-type learnng. 13 The results from Specfcaton 2 addtonally show that there are statstcally sgnfcant treatment dfferences between the two group lendng treatments ( χ 2 ( 2)= 31.24; p value = 0.00), but ths dfference s mnor durng the late perods (see fgure 3). The probablty of lendng s lower n the sessons conducted n Jadavpur. In summary (see Table 6), we fnd support for hypothess 1a (lendng rate s the lowest n the ndvdual lendng hgh cost treatment) and 1b (lendng rate s no hgher n the smultaneous group lendng treatment compared to the sequental group lendng treatment), but not for 1c (because the lendng rate s lower n the ndvdual lendng low cost treatment compared to the smultaneous group lendng treatment). Montorng Fgure 4 presents the average level of montorng across perods. Montorng rates are sgnfcantly lower n the ndvdual lendng hgh cost treatment compared to the low cost treatments (both ndvdual and group lendng). Controllng for montorng costs however, there s lttle dfference n montorng rates between ndvdual and group lendng (Table 2, Panel ). Montorng rates n the ndvdual lendng low cost treatment are sgnfcantly hgher compared to those n the group lendng treatments n the frst 5 perods, but ths dfference dsappears over tme.. The montorng decson s made by the lender n the ndvdual treatments and by a peer borrower n the group lendng treatments. 14 The level of montorng chosen s restrcted n the range [ 0,1 ] and s estmated usng a tobt model. 12 The relevant test here s 1 GROUP = 1 INDVLOWCOST and GROUP = INDVLOWCOST ;.e., both the slope t t 2 and the ntercept are dfferent. The test statstcs (dstrbuted as χ ( 2) under the null hypothess) are shown n the lower secton of Table Most of the demographc control varables are not statstcally sgnfcant n a consstent manner. Though we control for them n the regressons, we do not dscuss them n the text. Detals are avalable on request. 14 The propensty to make the loan s sgnfcantly lower n the ndvdual lendng treatments (partcularly n the hgh cost treatment), mplyng that the data on the level of montorng s often not observed n the case of the ndvdual lendng hgh 18

20 Consder frst the level of montorng chosen (by the lender) n the ndvdual lendng treatments. Table 4, Panel A, presents the random effects tobt regresson results and the Hausman- Taylor estmates for error component models. The ndvdual lendng low cost treatment dummy s postve and statstcally sgnfcant. Montorng rates fall over tme n both treatments and the sgn and sgnfcance of the ndvdual lendng low cost treatment dummy and the nteracton term wth 1 t ndcates that montorng rates are sgnfcantly lower n the ndvdual lendng hgh cost treatment. Addtonally the level of montorng n perod t 1 has a postve and statstcally sgnfcant relatonshp wth the level of montorng n perod t. As mentoned above n the case of group lendng (wth peer montorng), the payoff for an ndvdual borrower depends both on her level of montorng and also on the level of montorng of her partner. Subects could construct expectatons for the level of montorng of the other member of the group n dfferent ways. Here we consder the followng two smple alternatves: (1) Cournot expectatons: each subect expects the montorng level of the other member of the group to be the same as that n the prevous perod (Lagged Montorng of the Other orrower); (2) Fcttous play: each subect expects the montorng level of the other member of the group to be the average observed over all the prevous perods (Average Lagged Montorng of the Other orrower). Hence each subect s assumed to have a long memory as opposed to the Cournot expectatons case where each subect has a short memory. The results presented n Table 4, Panel show that montorng ncreased over tme and s modestly hgher wth sequental lendng (wth both specfcatons of expectaton formaton). Ths s consstent wth Hypothess 2b. The postve and sgnfcant coeffcent estmate of the other borrower s lagged montorng level (n the Cournot expectatons verson) or ts counterpart lagged average other borrower s montorng (n the fcttous play verson) s consstent wth the upwardly-sloped reacton cost treatment. The panel n ths case s therefore unbalanced: the observed number of montorng choces vares from 2 (.e., n only 2 of the possble 40 cases, dd the lender choose to make the loan) to

21 functons of the theoretcal model. Note that the coeffcent estmate on a borrower s own montorng n the prevous perod s also postve, and s substantally larger than the reacton to the other borrower s montorng level. Table 4, Panel C compares the level of montorng chosen across the lender and peer montorng treatments, holdng the cost of montorng constant. We present the results for two dfferent specfcatons: n specfcaton 1 we nclude a group lendng treatment dummy as defned above whle n specfcaton 2 we nclude separate dummes for the sequental and smultaneous group lendng treatments and the correspondng tme nteracton terms. The reference category n both cases s the ndvdual lendng low cost treatment. Specfcaton 2 n the random effects tobt regresson ndcates a sgnfcantly dfferent (upward) tme trend for group lendng, and overall the null hypothess of no dfference n montorng rates between the sequental group lendng treatment and the ndvdual lendng low cost treatment s reected. The results do not ndcate any evdence that the montorng rate s dfferent for the smultaneous group lendng and the ndvdual lendng low cost treatments, consstent wth hypothess 2c. Fnally combnng the results n Panels A, and C we fnd support for Hypothess 2a (montorng rate s the lowest n the ndvdual lendng hgh cost treatment). Repayment Rate The repayment rate s not a drect choce varable but s the result of a combnaton of the ex ante proect choce by the borrower, the level of montorng chosen by the borrower or lender, and the success of the montorng process: repayment occurs f the borrower chooses proect S or f the borrower chooses proect R and montorng s successful. Panel C of Table 2 shows that repayment rates, lke the other performance measures, are not sgnfcantly dfferent across the two group lendng treatments. Repayment rates are sgnfcantly lower n the ndvdual lendng hgh cost treatment compared to all three low montorng cost treatments. The average proporton of subects (ex ante) choosng proect R s sgnfcantly lower, however, n both the ndvdual lendng treatments compared to the group lendng treatments (Panel D of Table 2). 20

22 Table 5 presents random effect probt regresson results (wth the subect representng the random effect and the standard errors clustered at the sesson level) for repayment (columns 1 and 2) and ex ante choce of proect R (columns 3 and 4). The explanatory varables are smlar to the ones n Table 3 and as before we present the results from two alternatve specfcatons. The repayment rates are sgnfcantly lower n the ndvdual lendng hgh cost treatment compared to all other treatments (provdng support for Hypothess 3a). The results n Specfcaton 2 provde support for Hypotheses 3b and 3c (repayment rates are at least as hgh n the sequental group lendng and ndvdual lendng low cost treatments compared to the smultaneous group lendng treatment). Recall that the earnngs of the borrower are greater f he chooses proect R, but the earnngs of the lender are lower f the borrowers choose proect R. The non-parametrc tests reported n Panel D of Table 2 and the tests of ont sgnfcance reported n columns 3 and 4 of Table 5 ndcate that the borrowers are less lkely to choose proect R n the two ndvdual lendng treatments. 15 Table 4 earler showed that borrowers n these group lendng treatments are also more lkely to choose a hgh level of montorng to be able to swtch the other borrower s proect choce to S. Consequently, the actual proect choces are lkely to be proect S and the earnngs of the lenders are postve and outcomes move toward an effcent (montorng/lendng) equlbrum. On the other hand n the ndvdual lendng hgh cost treatment montorng rates are lower and even though borrowers are more lkely to choose proect S (.e., are less lkely to choose proect R compared to the theoretcal predcton), lenders choose not to make the loan. Outcomes n ths treatment frequently correspond to the neffcent (low montorng/no lendng) equlbrum. Fnally, holdng montorng cost constant the repayment rates are sgnfcantly hgher n the ndvdual lendng treatment compared to the smultaneous group lendng treatment. Snce montorng rates are not dfferent across these treatments (Table 4, Panel C), the 15 2 The correspondng value of the ( 2) χ statstc s (p-value = 0.00). 21

23 dfference s drven by the fact that borrowers are sgnfcantly more lkely to (ex ante) choose proect R n ths group lendng treatment compared to the ndvdual lendng treatment. One possble explanaton for the lower rate of choce of proect R n the two ndvdual lendng treatments could be that recprocal motvatons are trggered more n the two person ndvdual lendng game than the three person group lendng game. Lendng n the experment shares some parallels wth the frst move n the trust game (e.g., McCabe, Rgdon and Smth (2003)), and repayment and choce of the verfable proect s analogous to recprocal trustworthness. In the group lendng treatments, a par of borrowers may recprocate the lender s decson, but results from prevous trust game experments have shown that less recprocty s exhbted by groups (Song (2009)) or by ndvdual representatves decdng for groups (Song (2008)). In our group lendng envronment, subects appear to be less lkely to exhbt recprocal behavour when a fellow borrower s montorng, and ths may have reduced the borrower s perceved responsblty to be recprocal to the lender. These socal preference and recprocty concerns are not ncluded n theoretcal models of mcrofnance, such as the one motvatng our experment that assumes agents are own monetary payoff maxmsers. Our results add to the consderable expermental evdence (both n the lab and n the feld) that has accumulated n recent years ndcatng that ndvduals do not necessarly act as payoff maxmsers, and that other socal preferences often nfluence behavour (Sobel (2005)). Specfcally, t suggests that recprocal motvatons affect behavour more n the ndvdual lendng treatments than n the group lendng treatments. If addtonal evdence accumulates to ndcate that ths fndng s robust, t may be approprate to extend theoretcal models of mcrofnance to nclude socal preferences to mprove ther descrptve accuracy. Although observng socal preferences n the laboratory does not guarantee that they exst n the feld, we see no reason why such socal nfluences would not appear n actual mcrofnance relatonshps. 22

24 6. Implcatons of our Results and some Concludng Comments Our experment examnes several aspects of group lendng programs. The frst s the argument that sequental lendng s crucal to the success of group lendng schemes. We examne the emprcal valdty of theoretcal predctons regardng the added benefts of sequental lendng by comparng ts performance to smultaneous lendng n the presence of moral hazard and costly peer montorng, holdng constant mportant factors such as montorng costs. The second ssue s whether peer montorng ndeed does better than actve lender montorng. The lender s often an outsder who has less nformaton compared to peers about the borrowers. orrowers usually lve near each other and are more lkely to have closer socal tes. The thrd ssue s the relatve benefts of ndvdual and group lendng. Over the years there has been a dscernble shft from group lendng to ndvdual lendng n mcrofnance programs, and a number of theoretcal reasons have been advanced to explan ths shft. Frst, clents often dslke tensons caused by group lendng. Second, low qualty clents can free-rde on hgh qualty clents leadng to an ncrease n default rates. Thrd, group lendng can be more costly for the clents as they often end up repayng the loans of ther peers. Theoretcally the results are mxed. Our laboratory experment s able to address each of these ssues, through random assgnment of partcpants to group and ndvdual lendng treatments, and random assgnment to specfc lendng groups. We compare treatments when credt s provded to members of the group (sequentally or smultaneously) who can then montor each other, to a framework n whch loans are gven to ndvduals who are montored by the lenders drectly. Our results show that when montorng costs are lower for peer montorng than lender montorng, group lendng performs better compared to ndvdual lendng. Ths s reflected n hgher loan frequences and repayment rates. Ths occurs even though repayment rates wth ndvdual lendng consderably exceed the theoretcal predcton, whch mght reflect socal preferences such as recprocty. However f we hold the cost of montorng constant across the dfferent montorng regmes, then the performance of ndvdual and group lendng schemes are equvalent. Our fndngs therefore suggest an alternatve reason for the emergng popularty of 23

25 ndvdual lendng schemes, partally corroboratng the observatons of Gne and Karlan (2009) and Kono (2006). 16 Much of the success of mcrocredt programs has been attrbuted to self-selected groups and socal tes n rural communtes. However successful applcaton of these programs n other scenaros and economes requres more than ust strong socal tes. In urban contexts of developng and transtonal economes, for example, t mght be more dffcult to form self-selected borrowng groups. The optmal desgn of mcrocredt programs may need to look beyond the ssue of self-selecton and even look beyond group lendng. Indeed, expanson of mcrocredt and mcrofnance schemes to urban slums n developng countres could requre a dfferent approach. Socal captal and long term relatonshps between borrowers are vrtually non-exstent n urban slums. Ths suggests that a sgnfcant cost dfferental between lender and peer montorng s unlkely. Experments such as ths can exogenously manpulate montorng costs and forms of ndvdual and group lendng. If our results are robust to other envronments, they ndcate that ndvdual lendng programs can perform at least as well as group lendng programs. 16 In the frst experment reported n Gne and Karlan (2009), as the exstng feld centers wth group lablty loans were converted to ndvdual lablty loans, lenders had pror nformaton about the borrowers characterstcs from the group lendng feld sessons, whch could be used n the ndvdual lendng sessons at no extra cost. As a result the montorng costs dd not change substantally as they moved from group lendng to ndvdual lendng. Furthermore, partcpants had some experence wth group lendng before swtchng to ndvdual lendng. Ths suggests that montorng costs n that feld experment mght have been smlar under ndvdual lendng (wth actve lender montorng), compared to group lendng (wth peer montorng). Our laboratory experment results are consstent wth that nterpretaton. 24

26 Appendx A. Condton for the lender choosng to lend n the Smultaneous Lendng Model Recall that the lender s returns are: 2r 2 wth probablty mm ; H 2 wth probablty m ( 1 m ) ; H 2 wth probablty m ( 1 m ) ; and 2 wth probablty ( 1 m )( 1 m ). So the lender s expected earnngs are: mm ( 2r 2) + m ( 1 m )( H 2) + m ( 1 m)( H 2) + ( 1 m)( 1 m )( 2 ). The lender wll choose to lend as long as: mm ( 2r 2) + m ( 1 m )( H 2) + m ( 1 m)( H 2) + ( 1 m)( 1 m )( 2) > 2r. ** ** ** Snce borrowers are symmetrc and n equlbrum m = m = m, the lender wll lend f ( 2 2 ) 2 2( 1 ) ** 2 m r H m ** H r + > +, whch smplfes to the condton n the text: ( ) 2 ** ** m r H + m H > 1+ r. Appendx : Dervaton of Reacton Functons n the Sequental Lendng Treatment To obtan the reacton functons note that borrower earns: H r wth probablty mm f both borrowers choose proect S (.e., f both borrowers and are successful n the montorng process). 0 wth f probablty α ( 1 m) m (f s the frst borrower and s successful n the montorng process but s not) or wth probablty ( 1 α )( 1 m ) (f s the second borrower and s not successful n the montorng process). b wth probablty α ( 1 m )(f s the frst borrower and s not successful n the montorng process) or wth probablty ( 1 α )( 1 m ) m (f s the second borrower and s successful n the montorng process but s not). Appendx C: Condton for the lender choosng to make loans n Sequental Lendng Treatment Recall the lender earns: 2r 2 wth probablty mm (.e., both borrowers are successful n montorng); H 2 wth probablty ( ) (.e., only the second borrower s successful n montorng); and ( 1 r) 1 m m probablty ( 1 m ) (.e., the second borrower s not successful n montorng). So the expected return to the lender by choosng to lend s: mm ( 2r 2) + ( 1 m) m ( H 2) + ( 1 m )( 1+ r ). The lender wll choose to lend as long as ( ) ( ) ( ) ( )( ) + wth mm 2r m m H m 1+ r > 2r. Snce the borrowers are symmetrc and n equlbrum m = m = m, the lender wll lend f: 2 ( ) ( ) ( ) ( )( ) m 2r m m H m 1+ r > 2r. Ths smplfes to the condton shown n Proposton 3. 25

27 Table 1: Theoretcal Predctons, Parameter Values and Earnngs n the Dfferent Treatments Panel A: Theoretcal Predctons for Chosen Parameters Crteron Indvdual Lendng Hgh Cost Indvdual Lendng Low Cost Smultaneous Group Lendng Treatment: Sequental Group Lendng Treatment Treatment Treatment neffcent equlbrum/effcent equlbrum Make Loan No Yes No/Yes Yes Montorng Rate 0 1 0/1 1 (Ex ante) Proect Choce R R R/R R Panel : Parameter Values Parameter Indvdual Lendng Hgh Cost Treatment Indvdual Lendng Low Cost Treatment Smultaneous Group Lendng Treatment Sequental Group Lendng Treatment H b r λ r α Panel C: Earnngs, Smultaneous Group Lendng Treatment Actual proect choce of borrower 1 Actual proect choce of borrower 2 Earnngs of borrower 1 Earnngs of borrower 2 Earnngs of lender S S $1.75 C 1 $1.75 C 2 $2.50 S R $0.00 C 1 $2.50 C 2 $2.00 R S $2.50 C 1 $0.00 C 2 $2.00 R R $2.50 C 1 $2.50 C 2 -$2.00 No loan s provded $0.00 $0.00 $1.50 Panel D: Earnngs, Sequental Group Lendng Treatment Actual proect choce of the frst borrower Actual proect choce of the second borrower Earnngs of frst borrower Earnngs of second borrower Earnngs of lender S S $1.75 C 1 $1.75 C 2 $2.50 S R $0.00 C 1 $2.50 C 2 $2.00 R S $2.50 C 1 $0.00 C 2 -$0.25 R R $2.50 C 1 $0.00 C 2 -$0.25 No loan s provded $0.00 $0.00 $1.50 Panel E: Earnngs, Indvdual Lendng Actual proect choce of borrower Earnngs of borrower Earnngs of lender S $1.75 $1.25 C R $2.50 $1.00 C No loan s provded $0.00 $0.75 Note: C 1 and C 2 denote the montorng costs ncurred by borrowers 1 and 2 n the smultaneous and sequental group lendng treatments respectvely and ths cost depends on montorng m [0,1] and s gven by c( m) = m 2 2. C denotes the montorng cost ncurred by the lender, and ths cost depends on montorng m [0,1] and s gven by c( m) = λm 2 2; λ = 4.5n the ndvdual lendng hgh cost treatment and λ = 1 n the ndvdual lendng low cost treatment. S denotes the verfable proect choce, and R denotes the non-verfable proect choce. 26

28 Table 2: Selected Descrptve Statstcs Full Sample Frst 5 perods Last 5 perods Panel A. Average Proporton Makng Loans Indvdual Lendng Hgh Cost Treatment Indvdual Lendng Low Cost Treatment Smultaneous Group Lendng Treatment Sequental Group Lendng Treatment Group Lendng Treatments Rank sum Test Indvdual Lendng Hgh Cost = Indvdual Lendng Low Cost ** ** *** Indvdual Lendng Low Cost = Group Lendng * Indvdual Lendng Hgh Cost = Group Lendng *** ** *** Smultaneous Group Lendng = Sequental Group Lendng Smultaneous Group Lendng = Indvdual Lendng Low Cost * Sequental Group Lendng = Indvdual Lendng Low Cost * Panel. Average Level of Montorng Indvdual Lendng Hgh Cost Treatment Indvdual Lendng Low Cost Treatment Smultaneous Group Lendng Treatment Sequental Group Lendng Treatment Group Lendng Treatments Rank sum Test Indvdual Lendng Hgh Cost = Indvdual Lendng Low Cost *** *** *** Indvdual Lendng Low Cost = Group Lendng ** Indvdual Lendng Hgh Cost = Group Lendng *** *** Smultaneous Group Lendng = Sequental Group Lendng Smultaneous Group Lendng = Indvdual Lendng Low Cost Sequental Group Lendng = Indvdual Lendng Low Cost ** Panel C. Average Repayment Rates Indvdual Lendng Hgh Cost Treatment Indvdual Lendng Low Cost Treatment Smultaneous Group Lendng Treatment Sequental Group Lendng Treatment Group Lendng Treatments Rank sum Test Indvdual Lendng Hgh Cost = Indvdual Lendng Low Cost *** * *** Indvdual Lendng Low Cost = Group Lendng Indvdual Lendng Hgh Cost = Group Lendng ** *** Smultaneous Group Lendng = Sequental Group Lendng Smultaneous Group Lendng = Indvdual Lendng Low Cost Sequental Group Lendng = Indvdual Lendng Low Cost Panel D. Average Proporton Choosng the Non-Verfable Proect R Indvdual Lendng Hgh Cost Treatment Indvdual Lendng Low Cost Treatment Smultaneous Group Lendng Treatment Sequental Group Lendng Treatment Group Lendng Treatments Rank sum Test Indvdual Lendng Hgh Cost = Indvdual Lendng Low Cost * Indvdual Lendng Low Cost = Group Lendng ** 1.865* ** Indvdual Lendng Hgh Cost = Group Lendng *** *** Smultaneous Group Lendng = Sequental Group Lendng Smultaneous Group Lendng = Indvdual Lendng Low Cost * Sequental Group Lendng = Indvdual Lendng Low Cost 2.049** * * sgnfcant at 10%; ** sgnfcant at 5%; *** sgnfcant at 1% 27

29 Table 3: Random Effect Probt Regressons for Makng Loans Specfcaton 1 Specfcaton 2 1/t 1.894*** 1.939*** (0.471) (0.470) 1/t GROUP *** (0.717) 1/t INDVLOWCOST (0.792) (0.813) 1/t GROUP_SIMUL ** (0.828) 1/t GROUP_SEQUEN *** (0.856) Group Lendng Treatment (Dummy) 0.857*** (0.084) Smultaneous Lendng Treatment (Dummy) 1.419*** (0.124) Sequental Lendng Treatment (Dummy) 1.069*** (0.109) Indvdual Lendng Low Cost Treatment (Dummy) 0.315*** 0.483*** (0.104) (0.094) Negatve Earnngs n Prevous Perod (Dummy) *** *** (0.075) (0.073) Sesson at Jadavpur Unversty (Dummy) *** *** (0.069) (0.055) Constant 6.646*** 3.973*** (0.992) (0.678) Number of observatons Number of ndvduals Treatment Effects (Jont Sgnfcance): χ 2 ( 2) Group Lendng = Indvdual Lendng Hgh Cost *** Indvdual Lendng Low Cost = Indvdual Lendng Hgh Cost 16.64*** *** Smultaneous Lendng =Indvdual Lendng Hgh Cost *** Sequental Lendng = Indvdual Lendng Hgh Cost *** Group Lendng = Indvdual Lendng Low Cost 19.28*** Sequental Lendng = Smultaneous Lendng 31.24*** Sequental Lendng = Indvdual Lendng Low Cost 71.69*** Smultaneous Lendng = Indvdual Lendng Low Cost 18.60** Standard errors (clustered at the sesson level) are n parentheses * sgnfcant at 10%; ** sgnfcant at 5%; *** sgnfcant at 1% Regressons control for: proporton of correct answers n quz, age and age squared, gender, whether subect s usness/economcs/commerce maor, locaton of resdence when aged 15, year at unversty and prevous experence n terms of partcpaton n economc experments. 28

30 Table 4: Level of Montorng Chosen. Panel A: Indvdual Lendng (Lender Montorng) Random Effect Tobt Regresson Hausman-Taylor Estmaton for Error Component Models 1/t 0.187** 0.171** (0.083) (0.076) 1/t INDVLOWCOST (0.116) (0.106) Indvdual Lendng Low Cost Treatment (Dummy) 0.251*** 0.244*** (0.035) (0.053) Lagged Montorng 0.441*** 0.306*** (0.034) (0.029) Sesson at Jadavpur Unversty (Dummy) 0.085** 0.116** (0.034) (0.052) Constant (0.351) (3.262) Number of observatons Number of ndvduals Treatment Effect (Jont Sgnfcance): χ 2 ( 2) Indvdual Lendng Low Cost = Indvdual Lendng Hgh Cost 53.36*** 21.27*** Standard errors n parentheses * sgnfcant at 10%; ** sgnfcant at 5%; *** sgnfcant at 1%; Regressons control for: proporton of correct answers n quz, age and age squared, gender, whether subect s usness/economcs/commerce maor, locaton of resdence when aged 15, year at unversty and prevous experence n terms of partcpaton n economc experments. IV estmates to account for the possblty that the lagged dependent varable (lagged level of montorng) can be correlated wth the tme nvarant component of the error term (the unobserved ndvdual level random effect). Ignorng ths could result n based estmates. 29

31 Table 4 (contnued): Level of Montorng Chosen. Panel : Group Lendng (Peer Montorng) Random Effects Tobt Regresson Cournot elefs Hausman- Taylor Estmaton for Error Component Models Fcttous Play elefs Random Effects Tobt Regresson Hausman- Taylor Estmaton for Error Component Models 1/t (0.081) (0.062) (0.081) (0.062) 1/t GROUP_SEQUEN *** *** ** *** (0.110) (0.085) (0.111) (0.086) Sequental Lendng Treatment (Dummy) 0.054* 0.058** 0.053* 0.060** (0.029) (0.024) (0.030) (0.024) Lagged Own Montorng 0.504*** 0.349*** 0.496*** 0.340*** (0.022) (0.016) (0.022) (0.016) Lagged Montorng of the Other orrower 0.131*** 0.104*** (0.018) (0.014) Average Lagged Montorng of the Other 0.268*** 0.244*** orrower (0.056) (0.043) Sesson at Jadavpur Unversty (Dummy) ** (0.032) (0.028) (0.033) (0.029) Constant (0.944) (2.231) (0.960) (2.311) Number of observatons Number of ndvduals Treatment Effect (Jont Sgnfcance): χ 2 ( 2) Sequental Lendng = Smultaneous Lendng 8.07** 14.82*** 7.30** 13.53*** Standard errors n parentheses * sgnfcant at 10%; ** sgnfcant at 5%; *** sgnfcant at 1% Regressons control for: proporton of correct answers n quz, age and age squared, gender, whether subect s usness/economcs/commerce maor, locaton of resdence when aged 15, year at unversty and prevous experence n terms of partcpaton n economc experments. : See explanaton n Table 4 panel A. 30

32 Table 4 (contnued): Level of Montorng Chosen. Panel C: Comparng Peer Montorng and Lender Montorng wth Low Cost Random Effects Tobt Regresson Specfcaton 1 Specfcaton 2 Hausman-Taylor Estmaton for Error Component Models Specfcaton 1 Specfcaton 2 1/t (0.111) (0.111) (0.092) (0.090) 1/t GROUP ** ** (0.123) (0.102) Group Lendng Treatment (Dummy) 0.078** 0.108** (0.035) (0.044) 1/t GROUP_SIMUL (0.133) (0.107) 1/t GROUP_SEQUEN *** *** (0.135) (0.109) Smultaneous Lendng Treatment (Dummy) (0.038) (0.046) Sequental Lendng Treatment (Dummy) 0.104*** 0.112*** (0.037) (0.042) Lagged Own Montorng 0.515*** 0.510*** 0.357*** 0.354*** (0.020) (0.020) (0.015) (0.015) Sesson at Jadavpur Unversty (Dummy) * (0.029) (0.029) (0.038) (0.037) Constant * (0.692) (0.692) (6.254) (5.866) Number of observatons Number of ndvduals Treatment Effects (Jont Sgnfcance): χ 2 ( 2) Group Lendng = Indvdual Lendng Low Cost 7.00** 9.32** Smultaneous Lendng = Indvdual Lendng Low Cost Sequental Lendng = Indvdual Lendng Low Cost 13.63*** 18.23*** Sequental Lendng = Smultaneous Lendng 10.93*** 17.26*** Standard errors n parentheses. * sgnfcant at 10%; ** sgnfcant at 5%; *** sgnfcant at 1%. Regressons control for: proporton of correct answers n quz, age and age squared, gender, whether subect s usness/economcs/commerce maor, locaton of resdence when aged 15, year at unversty and prevous experence n terms of partcpaton n economc experments. : See explanaton n Table 4 panel A. 31

33 Table 5: Random Effect Probt Regressons for Repayment and Choce of Non-Verfable Proect (R) Repayment Choce of Proect R Specfcaton 1 Specfcaton 2 Specfcaton 1 Specfcaton 2 1/t (0.266) (0.265) (0.361) (0.365) 1/t GROUP *** (0.317) (0.412) 1/t INDVLOWCOST (0.349) (0.349) (0.487) (0.489) 1/t GROUP_SIMUL *** (0.372) (0.484) 1/t GROUP_SEQUEN *** (0.345) (0.438) Group Lendng Treatment (Dummy) 0.217*** 0.996*** (0.084) (0.136) Smultaneous Lendng Treatment (Dummy) *** (0.129) (0.101) Sequental Lendng Treatment (Dummy) 0.305*** 0.677*** (0.083) (0.081) Indvdual Lendng Low Cost Treatment (Dummy) 0.409*** 0.410*** 0.175* 0.549*** (0.100) (0.099) (0.101) (0.071) Sesson at Jadavpur Unversty (Dummy) ** (0.095) (0.097) (0.060) (0.053) Constant *** (1.547) (1.460) (1.629) (1.148) Number of observatons Number of ndvduals Treatment Effects (Jont Sgnfcance): χ 2 ( 2) Group Lendng = Indvdual Lendng Hgh Cost 7.86** 54.44*** Indvdual Lendng Low Cost = Indvdual Lendng 23.11*** 23.18*** 7.18** *** Hgh Cost Smultaneous Lendng = Indvdual Lendng Hgh *** Cost Sequental Lendng = Indvdual Lendng Hgh Cost 13.71*** 86.43*** Group Lendng = Indvdual Lendng Low Cost *** Sequental Lendng = Smultaneous Lendng *** Smultaneous Lendng = Indvdual Lendng Low 4.63* 20.68*** Cost Sequental Lendng = Indvdual Lendng Low Cost *** Standard errors (clustered at the sesson level) are n parentheses * sgnfcant at 10%; ** sgnfcant at 5%; *** sgnfcant at 1% Regressons control for: proporton of correct answers n quz, age and age squared, gender, whether subect s usness/economcs/commerce maor, locaton of resdence when aged 15, year at unversty and prevous experence n terms of partcpaton n economc experments. 32

34 Table 6: Summary of Results Hypotheses Descrpton Results + Implcatons H1: Lendng 1a ILHC < Mn{ILLC, SIM, SEQ} Supported Lendng rate s the lowest n the ndvdual lendng hgh cost treatment. 1b SIM SEQ Supported Lendng rate s modestly hgher n sequental group lendng treatment compared to smultaneous group lendng treatment. 1c SIM ILLC Not supported Lendng rate s lower n the ndvdual lendng low cost treatment compared to the smultaneous group lendng treatment. H2: Montorng 2a ILHC < Mn{ILLC, SIM, SEQ} Supported Montorng rate s lowest n the ndvdual lendng hgh cost treatment. 2b SIM SEQ Supported Montorng rate s modestly hgher n late perods n sequental group lendng treatment compared to smultaneous group lendng treatment. 2c SIM ILLC Supported Montorng rate s not sgnfcantly dfferent n the ndvdual lendng low cost treatment compared to the smultaneous group lendng treatment. H3: Repayment 3a ILHC < Mn{ILLC, SIM, SEQ} Supported Repayment rate s lowest n the ndvdual lendng hgh cost treatment. 3b SIM SEQ Supported Repayment rate s equvalent n the sequental and smultaneous group lendng treatments. 3c SIM ILLC Supported Repayment rate s not lower n the ndvdual lendng low cost and the smultaneous group lendng treatments. ILHC: Indvdual Lendng Hgh Cost Treatment ILLC: Indvdual Lendng Low Cost Treatment SIM: Smultaneous Group Treatment SEQ: Sequental Group Lendng Treatment + All results are consstent across the conservatve non-parametrc tests and the multvarate regressons. 33

35 Fgure 1: Reacton Functons n Smultaneous lendng. Note that reacton functons ntersect n two places (at (0, 0) and at (1, 1)), whch leads to multple equlbra. (0, 0) (1, 1) 34

36 Fgure 2: Reacton Functons of orrower n the Sequental Lendng Treatment 35

37 Fgure 3: Average Proporton Makng Loan, by treatment Average Proporton Makng Loan Perod Indvdual Lendng Hgh Cost Sequental Group Lendng Smultaneous Group Lendng Indvdual Lendng Low Cost 36

38 Fgure 4: Average Montorng Level, by Treatment Average Montorng Level Perod Indvdual Lendng Hgh Cost Sequental Group Lendng Smultaneous Group Lendng Indvdual Lendng Low Cost 37

39 References: [1] Abbnk, K.,. Irlenbusch and E. Renner (2006): "Group Sze and Socal Tes n Mcrofnance Insttutons", Economc Inqury, 44(4), [2] Anket, K. (2006): Sequental Group Lendng wth Moral Hazard, Mmeo London School of Economcs. [3] Armendarz de Aghon,. and C. Goller (2000): "Peer Group Formaton n an Adverse Selecton Model", Economc Journal, 110, [4] Armendarz de Aghon,. and J. Morduch (2000): "Mcrofnance beyond Group Lendng", Economcs of Transton,, 8, [5] Armendarz de Aghon,. and J. Morduch (2005): The Economcs of Mcrofnance, Cambrdge and London: MIT Press. [6] aneree, A., E. Duflo, R. Glennerster and C. Knnan (2009): The mracle of mcrofnance? Evdence from a randomzed evaluaton, MIT. [7] aneree, A. V., T. esley and T. W. Gunnane (1994): "Thy Neghbor s Keeper: The Desgn of a Credt Cooperatve wth Theory and a Test", Quarterly Journal of Economcs, 109, [8] arr, A. (2001): Socal Dlemmas and Shame-ased Sanctons: Expermental Results from Rural Zmbabwe, Mmeo, Centre for the Study of Afrcan Economes, Unversty of Oxford.. [9] ellman, E. (2006): Invsble Hand: Entrepreneur Gets g anks to ack Very Small Loans; Mcrolendng-for-Proft Effort In Inda Draws usness From Ctgroup, HSC; Ms. Dobbala's aby uffalo. Wall Street Journal, Eastern Edton. [10] hole,. and S. Ogden (2010): "Group lendng and ndvdual lendng wth strategc default", Journal of Development Economcs, 91, [11] Carpenter, J., S. owles and H. Gnts (2006): Mutual Montorng n Teams: Theory and Expermental Evdence on the Importance of Recprocty, IZA Dscusson Paper # [12] Cassar, A., L. Crowley and. Wydck (2007): "The effect of socal captal on group loan repayment: evdence from feld experments ", Economc Journal 117(517), F85 - F106. [13] Che, Y.-K. (2002): "Jont Lablty and Peer Montorng under Group Lendng", Contrbutons to Theoretcal Economcs, 2(1). [14] Chowdhury, P. R. (2005): "Group-lendng: Sequental Fnancng, Lender Montorng and Jont Lablty", Journal of Development Economcs, 77, [15] Conln, M. (1999): "Peer Group Mcro-Lendng Programs n Canada and the Unted States", Journal of Development Economcs, 60, [16] Fehr, E. and S. Gaechter (2000): "Cooperaton and Punshment n Publc Goods Experments", Amercan Economc Revew, 90(4), [17] Fegenberg,., E. Feld and R. Pande (2011): The Economc Returns to Socal Interacton: Expermental Evdence from Mcrofnance, Harvard Unversty. [18] Feld, E. and R. Pande (2008): "Repayment Frequency and Default n Mcrofnance: Evdence from Inda", Journal of the European Economc Assocaton, 6(2/3), [19] Fschbacher, U. (2007): "z-tree-zurch Toolbox for Readymade Economc Experments", Expermental Economcs, 10(2), [20] Fscher, G. (2008): Contract Structure, Rsk Sharng and Investment Choce, London School of Economcs, London.. [21] Fry, T. R. L., S. Mhalo, R. Russell and R. rooks (2008): "The factors nfluencng savng n a matched savngs program: Goals, knowledge of payment nstruments, and other behavor". Journal of Famly and Economc Issues, 29, [22] Ghatak, M. (2000): "Screenng by the Company You Keep: Jont Lablty Lendng and the Peer Selecton Effect", Economc Journal, 110, [23] Ghatak, M. and T. W. Gunnane (1999): "The Economcs of Lendng wth Jont Lablty", Journal of Development Economcs, 60, [24] Gne, X., P. Jakela, D. S. Karlan and J. Morduch (2010): "Mcrofnance Games", Amercan Economc Journal: Appled Economcs, 2(3), [25] Gne, X. and D. Karlan (2009): Group versus Indvdual Lablty: Long Term Evdence from Phlppne Mcrocredt Lendng Groups, Mmeo, Yale Unversty. [26] Hermes, N. and R. Lensnk (2007): "The Emprcs of Mcrofnance: What do we Know?", Economc Journal, 117, F1 - F10. [27] Karlan, D. (2007): "Socal Connectons and Group ankng", Economc Journal 117, F52 - F84. 38

40 [28] Karlan, D. and J. Znman (2010): "Expandng Credt Access: Usng Randomzed Supply Decsons to Estmate the Impacts", Revew of Fnancal Studes, 23(1), [29] Kono, H. (2006): Is Group Lendng a Good Enforcement Scheme for Achevng Hgh Repayment Rates? Evdence from Feld Experments n Vetnam, Mmeo, Insttute of Developng Economes, Chba, Japan.. [30] Masclet, D., C. Noussar, S. Tucker and M. C. Vlleval (2003): "Monetary and Nonmonetary Punshment n the Voluntary Contrbutons Mechansm", Amercan Economc Revew, 93(1), [31] McCabe, K., M. Rgdon and V. Smth (2003): "Postve Recprocty and Intentons n Trust Games", Journal of Economc ehavor and Organzaton, 52, [32] Morduch, J. (1999): "The Mcrofnance Promse", Journal of Economc Lterature, 37, [33] Ra, A. and T. Sostrom (2004): "Is Grameen Lendng Effcent? Repayment Incentves and Insurance n Vllage Economes", Revew of Economc Studes, 71(1), [34] Ray, D. (1998): Development Economcs, Prnceton Unversty Press. [35] Schellng, T. (1980): The Strategy of Conflct Cambrdge: Harvard Unversty Press. [36] Seddk, M. W. and M. Ayed (2005): Cooperaton and Punshment n Group Lendng, the Expermental Case, Mmeo. [37] Sobel, J. (2005): "Interdependent Preferences and Recprocty", Journal of Economc Lterature, 43, [38] Song, F. (2009): "Intergroup Trust and Recprocty n Strategc Interactons: Effects of Group Decson-Makng Mechansms", Organzatonal ehavor and Human Decson Processes, 108, [39] Song, F. (2008): "Trust and Recprocty ehavor and ehavoral Forcasts: Indvdual versus Group- Representatves", Games and Economc ehavor, 62, [40] Stgltz, J. (1990): "Peer Montorng and Credt Markets", World ank Economc Revew, 4, [41] Van Huyck, J., R. attalo and R. el (1990): "Tact Coordnaton Games, Strategc Uncertanty, and Coordnaton Falure", Amercan Economc Revew, 80, [42] Van Tassel, E. (1999): "Group Lendng Under Asymmetrc Informaton", Journal of Development Economcs, 60, [43] Varan, H. (1990): "Montorng Agents wth Other Agents", Journal of Insttutonal and Theoretcal Economcs, 146, [44] Walker, J. and M. Halloran (2004): "Rewards and Sanctons and the Provson of Publc Goods n One-Shot Settngs", Expermental Economcs, 7(3),

41 Appendx D: Instructons (Smultaneous Lendng Treatment) General: Ths s an experment n the economcs of decson-makng. The nstructons are smple and f you follow them carefully and make good decsons you wll earn money that wll be pad to you prvately n cash at the end of the expermental sesson. Your earnngs wll be n expermental dollars and they wll be converted nto real dollars at the followng rate: 1 Expermental Dollar = Real Dollars. Notce that you earn more money by earnng more expermental dollars. After we fnsh readng the nstructons and before we start the experment, we would lke you to answer a set of questons relatng to these nstructons. You wll be pad n cash (at the end of the experment, n addton to your earnngs from the actual experment) at the rate of $0.50 for each correct answer. In today s experment, you wll be randomly dvded nto groups and each group wll have three members. Each group conssts of one lender and two borrowers. Your role ether borrower or lender s determned randomly and wll reman unchanged throughout the experment. At the end of every perod, partcpants wll be randomly re-matched and so the other people n your group wll typcally change each perod. You wll make decsons for 40 perods. Decson Makng: Two proects are avalable to each borrower every perod: proect S and proect R. The cost of each proect s $1 and t s to be fnanced by a loan from the lender. Every perod the lender can choose whether or not to nvest her $2 nto makng loans to the borrowers. She must ether make the loan to both borrowers or to nether borrower, and she cannot make a loan to a sngle borrower. If the lender chooses not to nvest n the loans to the borrowers, she earns $1.50 for the perod. If the borrowers receve the loans, they can montor the proect choce of the other borrower n ther group by choosng to pay a montorng cost (C). oth borrowers can montor each other. If borrower X ncurs a cost C on montorng, there s a chance of M that the other borrower Y wll automatcally be requred to choose proect S. Otherwse the other borrower can choose ether proect S or proect R. Choces wll be made smultaneously and the borrowers wll not know whether the lender chooses to make the loans or not before makng ther choce of proect. All decsons wll be revealed after both the lender and the borrowers have made ther decsons. orrowers pay ther selected montorng costs whenever the lender makes the loan, regardless of whether or not the montorng s successful. The montorng chances work n the followng way. Suppose borrower X chooses M = 20%. In ths case, magne an urn (or the bngo cage the expermenter s holdng) contanng 10 total balls: 2 whte balls and 8 red balls. One ball s drawn from ths magnary urn, and f we draw a whte ball then a borrower Y choce of R s swtched to S; f we draw a red ball then a borrower Y choce of R remans R. If borrower Y chose S, then ths choce of S s mplemented regardless of the ball draw. Remember, borrower Y also makes montorng choces n the same way to possbly swtch borrower X choces from R to S. 1

42 To take another example, suppose borrower Y chooses M = 70%. In ths case you should magne an urn contanng 7 whte balls and 3 red balls. Agan, a drawn whte ball swtches a borrower X choce of R to S, but a drawn red ball means that a borrower X choce of R remans R. Therefore, a hgher choce of M, whch s more costly as shown n the table below, ncreases the chances that the other borrower s choce of R s swtched to S. A dfferent ball draw, from a dfferent magnary urn, s conducted for every dfferent group and borrower for every dfferent perod n the experment. In other words, the random draws are all ndependent. The relatonshp between C and M s as follows: Montorng Cost (C) M Interpretaton of M percentage: $ % Swtch a borrower choce of R to S 0 out of 10 tmes $ % Swtch a borrower choce of R to S 1 out of 10 tmes on average $ % Swtch a borrower choce of R to S 2 out of 10 tmes on average $ % Swtch a borrower choce of R to S 3 out of 10 tmes on average $ % Swtch a borrower choce of R to S 4 out of 10 tmes on average $ % Swtch a borrower choce of R to S 5 out of 10 tmes on average $ % Swtch a borrower choce of R to S 6 out of 10 tmes on average $ % Swtch a borrower choce of R to S 7 out of 10 tmes on average $ % Swtch a borrower choce of R to S 8 out of 10 tmes on average $ % Swtch a borrower choce of R to S 9 out of 10 tmes on average $ % Swtch a borrower choce of R to S 10 out of 10 tmes Earnngs: If they receve the loan, the earnngs of the borrowers depend on the proect choces made by the two borrowers and on the montorng costs the two borrowers choose to ncur. If the lender decdes to make the loan, her earnngs depend on the actual proect choces made by the two borrowers. If she chooses not to nvest n the loans to the borrowers, her money s allocated to a savngs account and she earns $1.50 for the perod. The earnngs of the two borrowers and the lender n the dfferent proect scenaros are as follows. Here C 1 and C 2 denote the montorng costs ncurred by borrower 1 and 2 respectvely. Actual proect choce of borrower 1 Actual proect choce of borrower 2 Earnngs of borrower 1 Earnngs of borrower 2 Earnngs of lender S S $1.75 C 1 $1.75 C 2 $2.50 S R $0.00 C 1 $2.50 C 2 $2.00 R S $2.50 C 1 $0.00 C 2 $2.00 R R $2.50 C 1 $2.50 C 2 -$2.00 No loan s provded $0.00 $0.00 $1.50 Each borrower can ncrease the chances of the other choosng proect S by nvestng n montorng. Montorng choces wll have to be made smultaneously and before each borrower knows whether the lender actually makes the loan. Examples: 2

43 Consder the followng examples, whch were chosen randomly and are not meant to suggest any partcular decsons. Example # 1: 1. Lender makes the loan. 2. orrower 1 chooses proect S and montorng M of 70%. Montorng cost C 1 = $ orrower 2 chooses proect R and montorng M of 30%. Montorng cost C 2 = $ Montorng results: orrower 1 s montorng s unsuccessful and so borrower 2 s actual proect s proect R. orrower 2 s montorng s also unsuccessful, but borrower 1 already chose proect S, and so hs actual choce remans proect S. 5. Earnngs: Use the second row of the prevous table to determne borrower 1 s earnng = $( ) = $ 0.245; borrower 2 s earnng = $( ) = $2.445; and lender earnng = $2.00 Example # 2: 1. Lender makes the loan. 2. orrower 1 chooses proect R and montorng M of 80%. Montorng cost C 1 = $ orrower 2 chooses proect S and montorng M of 50%. Montorng cost C 2 = $ Montorng results: orrower 1 s montorng s successful, but borrower 2 already chose proect S and so hs actual proect choce remans proect S. orrower 2 s montorng s also successful and ths swtches borrower 1 s actual proect choce to S. 5. Earnngs: Use the frst row of the prevous table to determne borrower 1 s earnng = $( ) = $1.430; borrower 2 s earnng = $( ) = $1.625; and lender earnng = $2.50. Summary of Decsons to be taken: Lender: 1. In every perod choose how you want to nvest your $2, usng a decson screen shown n Fgure 1. orrowers: 1. Indcate how much you wsh to nvest n montorng the other borrower to possbly swtch hm or her to proect S, as shown n Fgure 2, n case you receve the loans. 2. Decde whether you want to nvest n proect S or proect R, usng a decson screen shown n Fgure 3. Remember that f the other borrower chooses to ncur a montorng cost (shown as other s C on the fgure), there s a chance of M that your proect choce wll be swtched to S, even f you had actually chosen R. Remember that choces are made smultaneously and the borrowers do not know whether the lender chose to nvest n the loans or not before makng ther choces of a proect. Once both the lender and the borrowers have made ther decsons, the nformaton shown n Fgure 4 wll be provded to all of the partcpants n the group: orrower s proect choces Dd the lender choose to make the loan orrower s montorng level, f the lender chose to make the loan Actual proects chosen by the borrower, f the lender chose to make the loan Lender earnngs 3

44 orrower earnngs Your cumulatve earnngs over the experment Attached to these nstructons s a record sheet where you are requred to record your earnngs and other detals from every perod. Are there any questons before we start the experment? 4

45 Fgure 1: Lender s loan decson screen 5

46 Fgure 2: orrower s montorng decson screen 6

47 Fgure 3: orrower s proect decson screen 7

48 Fgure 4: Example output screen 8