The Invisible Handshake in Contracting for Debt: Results from an International Experiment
Alexander W. Cappelen, Rune Jansen Hagen, Erik Ø. Sørensen, and Bertil Tungodden
yHighly preliminary - please do not quote
Abstract
We conducted an experiment in Norway and Tanzania to study the behaviour of lenders and borrowers in situations where contracts cannot be enforced. In addition to the international dimension, our version of the Trust game entailed three changes compared to standard one. Firstly, attached to Lenders’o¤ers were one of two types of contracts: a standard
…xed interest rate contract where the Borrower’s obligation is independent of the outcome of the project or a surplus-sharing contract where the re- payment is contingent on the realisation of the investment that the funds borrowed was used for. Secondly, Borrowers had to explicitly accept or reject the loans. Thirdly, the investment outcome was uncertain. Prelim- inary analysis of the data suggests that equality, need, and contracts all matter for behaviour. However, the attraction of equality and the norm of ful…lling the obligations in contracts one has freely entered into clearly dominate the concern for the needs of your contractual partners.
1 Introduction
Millions of economic transactions are made every day in the modern economy, and the wealth of nations is a product of the gains from trade between individ- uals and …rms. As Adam Smith emphasised, the gains from trade are limited
We appreciate the excellent research assistance of Jofrey Amanyise, Ben Dandi, Øystein V. Dvergsdal, Lars Engebredtsen, Maria Frengstad, Ole-Aleksander Greve, Cornel Jahari, Lucas Katera, Hilda Luusah, Khadijah Omer, Dennis Rweyemamu, Stein Svalestad, Sander van den Heuvel, and Jon Vassengen. We are also grateful to REPOA (Dar-Es-Salaam) and the University of Bergen for letting us use their laboratories. Finally, we would like to thank participants at the 3rd Nordic Conference on Behavioral and Experimental Economics, Uni- versity of Copenhagen for comments. The research reported here has been …nanced by the Research Council of Norway and ESOP.
yCappelen: Norwegian School of Economics and Business Administration, email:
[email protected]; Hagen: University of Bergen, email: [email protected];
Sørensen: VU University Amsterdam, email: [email protected]; Tungodden: Norwegian School of Economics and Business Administration, email: [email protected].
by the extent of the market. A basic problem confronting all societies is how to nurture exchange by circumscribing the possibilities for cheating by parties to a transaction. Historical studies show how groups devised strategies that allowed them to move beyond the con…nes of simultaneous exchange. However, groups that succeeded in constructing impersonal institutions regulating the conduct of transactions ‡ourished while those that relied on personal ties to support trade across space and time eventually ran into natural constraints posed by the size of social networks (see e.g. Greif 2006). Today, international and intertemporal trade is an integral part of the economies of the rich countries and contributes signi…cantly to their citizens’ wealth and welfare. Indeed, the rapid growth of production and external trade after 1945 probably contributed to creating a form of “tunnel vision": for a long time the existence of the institutions nec- essary to sustain large, impersonal market economies was taken for granted by economists. In the Walrasian paradigm such institutions were merely part of the scenery and economists focussed on minor obstacles on the road to the perfect market economy such as imperfect competition. Complementing this view was the assumption that if the road got bumpy, individuals and …rms would simply meet at the wayside to iron out an e¢ cient agreement in Coasian fashion.
Today, economists habitually emphasise the importance of contracts and of formal juridical institutions in enforcing them. The complete contracting frame- work has been replaced by a body of work stressing the fact that contracts are always incomplete. Most basically, this is due to the impossibility of foresee- ing all future contingencies that could be relevant to the parties to a contract.
However, the transaction costs associated with the writing of formal agreements are recognised as well. Transactions might be precluded by problems relating to measuring and verifying the characteristics of the good or service to be ex- changed. Juridical institutions also shape patterns of exchange by in‡uencing the costs of third-party enforcement of contracts. This line of reasoning has been used to analyse the boundaries between …rms and markets, for example, as well as the optimal assignment of ownership and control rights in assets. The main cost of incomplete contracts has typically been judged to be the reduction of relation-speci…c investments, as parties to an agreement foresees a hold-up problem during ex post renegotiations.
However, the implication of this approach in cases where contracts are too costly to write or enforce is not necessarily that actors will everywhere and always try to grab as much possible of the gains from trade for the self, with the consequent loss of potentially mutually bene…cial trade. If they expect to interact again, individuals might be willing to foresake current gains for future ones. That is, they might be able to establish "implicit contracts" based on reputation, in which cooperation now fosters reciprocal cooperation the next time around. In such cases, an "invisible handshake" governs an agreement to make an e¤ort, refrain from taking advantage of the other parties to the contract, etc.
It is well-known that the requirements for such reputational equilibria to exist can be quite demanding in theory. For example, while the strategy of "tit- for-tat" might foster cooperation in the in…nitely repeated Prisoners’Dilemma,
this is only one of many equilibria and require individuals to value the future su¢ ciently. Still, in the real world there are many examples of situations were the loss of reputation and exclusion from future exchanges (or similar sanctions) sustain some degree of cooperation. They can be found at the local level, in the "village economies" of poor countries where interaction over generations enable villagers to provide mutual insurance and local public goods; or at the international level, where most governments are able to borrow even though there are no courts in which to enforce such contracts.
It is increasingly recognised that the stringent requirements for cooperation that standard game theory highlights are based on a fairly narrow view of the world. There is extensive evidence from most social sciences that humans are not only driven by a concern for their own material self-interest. For example, numerous experiments document how individuals behave according to norms that not only circumscribe their own opportunistic behaviour, but punish such acts by others even when it is not in the narrow self-interest of the punisher to do so.1 Obviously, if people does not take advantage of each others to the extent implied by standard game theory, many more transactions will be conducted even when contracts are necessarily incomplete and a functioning framework for enforcement is absent. Similarly, if cheating and opportunism are deterred by various forms of sanctions (social or individual) the moral economy is likely to be more prosperous than the one-eyed game theorist would have expected. A warm "invisible handshake" is thus probably part of the story of how the mod- ern economy functions relatively smoothly despite these signi…cant institutional impediments.
Still, there seems to be some limits to the strength of the moral economy. The
"village economies" just mentioned su¤er from the same problems as pre-modern societies: the gains from trade are heavily circumscribed by the need to limit the size of the network to a level where free-riding and other forms of opportunism can be contained. This leaves them vulnerable to common shocks and precludes them from fully realising the bene…ts of specialisation. Social norms can also be counterproductive. There are many examples of norms against wealth accumu- lation in such societies that stops individuals from opting out of a group, which obviously hinders economic growth. Similarly, sharing norms can be bene…cial for purposes of mutual insurance but detrimental to entrepreneurship and in- vestment by acting as a tax on such e¤orts. "Gentleman’s agreements" allowing for "excusable default" on sovereign debt in the event of unforeseen hardships for the borrower does not extend to the poorest countries in the world, which are mainly forced to rely on foreign aid instead of foreign loans. Nor do they prevent the spectatcular debt crises that hit sovereign borrowers from time to time. While the economic di¢ culties, including a temporary halt in the access to foreign capital, that follow defaults could be part of a reputational equilibrium under uncertainty, they indicate the potential bene…ts to be had from having adequate third-party enforcement of contracts. Indeed, the literature on insti- tutions and development …nd a strong positive correlation between the formal
1Camerer (2003) provides a nice overview of the literature.
protection of property and contract rights and the level of income a country has.2 While the direction of causality is still under debate this does suggest that impersonal institutions provide economic bene…ts beyond what informal institutions such as norms do.
We seek to probe the limits of the moral economy by conducting an exper- iment in two countries, Norway and Tanzania, simulating an environment in which contracts are not enforceable.
2 Design of the Experiment
2.1 Outline
We conducted a Trust Game with some twists to it in six sessions staged si- multaneously in two countries, Norway and Tanzania. In the standard Trust Game, one player decides how much of a given endowment her or she would like to entrust to a second, anonymous player. This sum is then multiplied by some factor so that the second player receives a larger sum than the …rst player sent. He or she then decides whether they want to return some of the money to the …rst player. In a sense, in this game someone with capital is paired with someone with a project and the question is whether the former trusts the latter enough to provide the money needed to get the project o¤ the ground.
In addition to the international dimension, our version of the game entailed three changes compared to the bare-bones one. Firstly, the …rst player (here- after: the Lender) was not only told to determine whether to risk some of the endowment by lending money to the second player (the Borrower), but was also informed of the details of the contract that could be o¤ered. We allowed for two di¤erent types of contracts. The …rst was a standard …xed interest rate contract where the Borrower’s obligation is independent of the outcome of the project.
The second was a surplus-sharing contract. The contract then speci…ed that the Borrower had to share the surplus from the project (the total returns minus the invested amount) with the Lender.
The second change we made was to make the Borrower explicitly accept or reject the contract on o¤er. That is, instead of just automatically transferring any money made available by the …rst player to the second after mechanically magnifying the amount by a certain factor, the Borrower had to decide whether to take on the obligations speci…ed in the contract before knowing what the returns to the project were. While we stressed that these obligations were not enforceable in any way, and so should not matter in a one-shot game of anony- mous interaction between rational, sel…sh individuals, there are many reasons why the explicit acceptance of a contract on certain terms might a¤ect behav- iour.
Thirdly, the returns to the project was in our speci…cation uncertain. That is, instead of having the amount lent multiplied by a given factor, the multiplier
2See e.g. Easterly and Levine (2003) for both original results and references to the litera- ture.
in our game could take on one of three randomly determined values: one, two, or three. These three values were equally likely to occur. The interest rate in the
…xed interest rate contract and the rate of surplus sharing were both set at 50%
so that the expected value of borrowing an amount B and repaying according to the contract was the same for the two types of contracts:
1 3B+1
32B+1
33B (1 +r)B = (1 r)B;
1
3(1 s) (B B) +1
3(1 s) (2B B) +1
3(1 s) (3B B) = (1 s)B:
We will now describe the experiment in more detail, starting with the lending phase.
2.2 The Lending Phase
At the start of the experiment, each participant was assigned a role as Lender or Borrower by the computer. Each Lender was given an endowment of 50 US dollars. Lenders were in the majority of cases given four decisions to make, where each decision was characterised by the nationality of the Borrower (Norwegian or Tanzanian) and the type of contract (Fixed Interest or Surplus Sharing).
The assignment of Borrower nationality and contract type was in each case determined randomly by the computer. As all participants had been informed that the contracts were not enforceable, Lenders knew that their money was at the mercy of the Borrowers if they decided to lend. They had the option of keeping their endowment, i.e., to refrain from passing any of it on to one or more Borrowers, in which case they would leave the experiment with 50 US dollars plus their show-up fee.3 They were also informed that only one of the decisions they were involved in would determine their …nal payment from the experiment if they decided to venture some of their money. However, in each situation they could lend up to their endowment, in multiples of 10 US dollars.
Once they had made all of their decisions, Lenders were given the option of revising them, i.e., to change the amount lent. They could thus decide to take their chances if their preliminary choice had been to lend zero dollars, or scale up and down their loan o¤ers within the con…nes of the 0-50 US dollar range.
When they had …nalised their decisions, Lenders had to wait until Borrowers had made theirs to know how they had fared. In the meantime they were presented with a set of questions pertaining to their demographic pro…le (age, gender, etc.), political views, and their opinions on issues of sovereign debt. They were also asked to write a text motivating their decisions in the experiment.
2.3 The Borrowing Phase
Those who were assigned the role of Borrower had to wait until all Lenders (in Norway as well as in Tanzania) had completed their decision-making. They were
3The show-up fee was XXX TZS in Tanzania and 100 NOK in Norway.
then presented with randomly chosen o¤ers, in most cases four, and had to make up their minds as to whether they would like to accept them. Just like Lenders, they were made aware of the nationality of their contractual counterpart and the type of contract applying. They were of course also informed of the size of the loan o¤ered. And like the Lenders they could revise their preliminary decisions once they had made them, thus for example switching from rejection to acceptance (or vice versa) if they so wished. Borrowers were aware that only one of the situations they confronted would be randomly drawn by the computer to determine their …nal payment from the experiment. Their options were also similar to those of Lenders in the sense that they too could choose to stay out of the market altogether by rejecting all o¤ers, in which case they would keep their endowment of 50 US dollars and receive the show-up fee as well.
2.4 The Repayment Decision
Once a Borrower had con…rmed all of his or her choices, the computer deter- mined the investment outcome in the manner described above. A Borrower would then be shown a screenshot summarising the …rst interaction they had been involved in. They were reminded of the nationality of the Lender, the type of contract on o¤er, and the size of the loan and informed about the investment outcome and the corresponding repayment according to the contract. Borrow- ers could then freely determine how much to repay in multiples of 5 US dollars within the range determined by the investment outcome. That is, they could repay anything from zero to the investment outcome. After they had made all their repayment choices they were given the option of revising their preliminary decisions.4
After Borrowers had …nally con…rmed their repayment choices, they were presented with the same questions as Lenders. They were also asked to motivate their choices.
3 Descriptive Statistics and Preliminary Results
3.1 The Participants
A total of 241 participants from the universities of Dar-es-Salaam, Tanzania and Bergen, Norway participated in the experiment, which was run simultaneously in the two countries for a total of six sessions per country. Each participant was recruited from the general population of university students at the bachelor level, and only participated once. The lowest (highest) number of participants in a session was 35 (42), with the di¤erence in the number of participants in each country never exceeding 2 in any session (c.f. Table 1).
[Table 1 about here]
4If a Borrower had rejected a loan o¤er, he or she were at this stage reminded that they had done so and that consequently there was no decision to make. They could not revise such a choice after the borrowing phase had been completed.
Overall, participation was fairly balanced in terms of nationality and gen- der; 51.0% of the participants were Tanzanians and 48.1% were male. In both countries, more than 80% of the participants were no older than 25 years. With respect to year of study, third-year students were somewhat overrepresented at 41.5% of the total while only 12.4% were in their fourth year. As Table 2 shows, the demographic characteristics of the participants are very similar in the two countries.
[Table 2 about here]
Table 3 demonstrates that in terms of the roles in the experiment that were assigned to participants, there are no signi…cant imbalances in terms of nation- ality and gender. The only di¤erence of note is that in Norway, females were chosen by the computer to be Lender in somewhat greater proportion than men.
[Table 3 about here]
3.2 A First Look at Lending Decisions
In Table 4, the choices of Lenders’ are recorded. In total, 472 loan decisions were made during the six sessions. Only 6.57% of these accorded with standard game theory in the sense that lending nothing is optimal if you believe that Borrowers will be opportunistic and decline to honor the contract. With no third-party enforcement available lending money is a losing proposition or, in a somewhat di¤erent light, a pure gift to Borrowers unless Lenders expect them to be guided by motivations other than pure material self-interest. There were in fact some participants that reasoned just like a Lender should according to standard game theory, i.e., by working backwards from the Borrowers’decision assuming that he or she would be opportunistic:
I decided not to lend away any money because there was no guarantee that
"my" borrower would pay me back. In fact; if I was a borrower; I would lend as much as I could and then not pay the money back. I would do this because the so-called contract could not be enforced in any way, and also because everyone involved in the transaction would remain anonymous. So I guess my decision not to lend away any money was based on me putting myself in the borrowers situation and concluding that - because that’s how I would have acted - this person would most likely be "sel…sh". (Motivation 27)
[Table 4 about here]
On average, Lenders o¤ered Borrowers 29.66 US Dollars, or almost exactly 60% of their endowments. Hence, they seemed to put a signi…cant amount of trust in Borrowers willingness to reciprocate their generosity or were perhaps to a signi…cant degree swayed by perceptions of need on the part of Borrowers.
Beyond these observations, the data shows a surprising regularity: about 1/5 of the loans are for 1/5 of Lenders’endowments (or less), in 2/5 of the loans Lenders venture 2/5 of their capital (or less), etc. That is, roughly speaking each feasible loan amount was made in about 20% of the decisions. This indicates that there might be some regularities in terms of the perceived motivations of potential Borrowers, and thus in the extent to which they could be trusted to return the funds with (some) interest, or in the extent to which they were
perceived to be in need so that Lenders are willing to donate money to them.
In the motivations provided by Lenders, one can …nd examples of this sort of complex reasoning:
I thought I’d help out the borrowers and maybe they’d help me out too! I hope and think that when someone borrows money he/she feels, at least to a certain degree, obligated to pay the money back. (Motivation 226)
First of all, the fact that the loan contract cannot be enforced makes lending a serious risk for the lenders. Since there is risk involved, i would never lend away all of my money. Since I cannot meet the person i’m going to lend the money to, i chose not to di¤ er between the di¤ erent types of loan, and where the borrower comes from. If i had any guarantee that the loan contract would be enforced, i would be more willing to lend money. (Motivation 238)
The cumulative distributions of loan o¤ers are not too di¤erent when dif- ferentiated by country. However, the distribution is somewhat more polarized in Norway, with 10.68% and 24.79% at the lowest and highest feasible level, respectively, compared to corresponding numbers of 2.52% and 18.07% for Tan- zania. This could indicate, …rst of all, that a higher proportion of Norwegians are sel…sh and thus expect others to behave opportunistically come repayment time and/or that, being more unfamiliar with environments in which the rule of law cannot be trusted to deliver contract full…lment, they are somewhat more averse to the risks involved.
The higher proportions of o¤ers involving the whole of the Lenders’ en- dowment is most likely due to the Norwegians perceiving need to be higher in Tanzania. That is, there could be a signi…cant gift element in these maximum o¤ers. This line of reasoning is supported by the fact that whereas only 22.03%
of the loans from a Norwegian Lender to a Norwegian Borrower was for 50 US dollars, 27.59% of the o¤ers from Norwegian Lenders to Tanzanian Borrowers were this high.
Figure 1 illustrates in more detail what the distribution of loan o¤ers to Bor- rowers of di¤erent nationalities looks like when Lenders are sorted by nationality too.
[Figure 1 about here]
Did the choices of Lenders depend on the type of contract they could o¤er?
The correlation between the amount lent and the type of contract is positive, implying that more money tend to be o¤ered when the contract on hand is of the surplus sharing type. However, the correlations are very small and nowhere near statistical signi…cance in neither the whole sample nor in the subsamples of Lenders of the same nationality. We had expected that there would be less lending with …xed interest rates as Lenders would surmise that if the investment went awry Borrowers would have to dig into their endowment in order to full…l the contract, which could be seen as a signi…cant sacri…ce (a variety of the
"Endowment E¤ect"). Apparently, Lenders were not thinking along these lines.
3.3 A First Look at Acceptance Decisions
As already noted, in the standard Trust Game any money provided to the trustee are automatically invested. We introduced an explicit need for accepting such transfers, and Table 5 shows that it matters. We were expecting it to matter because of a kind of commitment e¤ect: if you have explicitly accepted speci…c obligations towards another person it is probably harder to disregard them ex post. However, the table indicates that there might be another commitment e¤ect at work: a surprising number of potential Borrowers refused the loans o¤ered. As there is only upside risk in material terms to accepting loans, one explanation is that individuals know that they will not be able to resist the temptation of cheating ex post and thefore eschew the possibility of being in a situation where one acts contrary to one’s moral norms.
[Table 5 about here]
Table 6 illustrates that there are more Tanzanians than Norwegians that forego the opportunity to bene…t from the loan o¤ers actually made. Following the line of reasoning above one would expect that weakness of will is a greater problem when there is greater need, and as Tanzanians are poorer than Norwe- gians on average they would then be more inclined to commit themselves even if the proportions of people in the population having norms they fear they might violate in moments of weakness are the same.
[Table 6 about here]
Table 7 shows that the type of contract in question does not seem to matter much for the decision of whether to accept or reject an o¤er. This holds for Borrowers from both countries.
[Table 7 about here]
One might suspect that the acceptance rate could be a function of the size of the loan, as the amount of money on o¤er could be interpreted as a signal of the degree of trust in the Borrower. Table 8 suggest that if there is such an e¤ect, it is not monotonic. The very low acceptance rate for o¤ers of 10 US dollars could indicate that such small loans are seen as derisory, though.
[Table 8 about here]
3.4 A First Look at Repayment Decisions
Excluding the 31 cases were no loan was made and the 131 cases where the o¤er was not accepted, there are 310 operative loan contracts in our sample. 50.97%
of these were …xed interest rate contracts, demonstrating that the decisions by some participants not to make loans or not to accept o¤ers had no impact on the distribution of contract types. The formal implications of agreeing to a surplus sharing contract were not clear until the uncertainty about the returns to investment had been revealed. Ex post, some sizeable contractual obligations existed, the largest one being two times the endowment of Borrowers (i.e., 100 US dollars). The average obligation was 49.97 US dollars.
On average, Borrowers repayed 60.65% of their contractual obligations. This may of course be seen as a cup half full or an half empty one. The shortfall
of 40% is sizeable when viewed from a juridical angle, indicating that contract enforcement matters and that formal institutions therefore have a role to play.5 On the other hand, it is remarkable that people repay such a large share of their obligations when there is no penalty whatsoever associated with default. Within any reasonable margin of error, this cannot be explained by the standard theory in which indviduals only care about their own monetary gains.
From a di¤erent angle, Borrowers only returned 24.88% of the resources at their disposal, de…ned as the gross returns to investment plus their endowment, to Lenders. In other words, they kept 75% of the money already in their pocket.
This looks more in line with the standard game theoretic model of materially self-interested individuals at …rst sight, but in judging this one should keep in mind that Lenders have not necessarily lent all of their money. Thus, giving away 50% of the resources at their disposal would on average make Borrowers poorer than Lenders within the con…nes of the experimental setting. In fact, the average return share needed to create perfect equality between Borrowers and Lenders in terms of the total resources available was 40.31%. While Borrowers still fell short of this ideal, the margin is much smaller than the impression created by the raw return shares.
In terms of means, there are for the most part no great di¤erences if one controls for the nationalities of Lenders and Borrower, contract type, or the extent of luck. Somewhat surprisingly, though, the average return share when the investment multiplier is three is four percentage points below the average return share for multipliers of one and two.
In terms of correlations, the return share is only signi…cantly associated with the shares needed to generate equality and to full…l contractual obligations as well as the degree of luck. In line with the impression given by the averages the latter correlation is negative, indicating perhaps that Borrowers are good at rationalising the extent to which they deserve good luck.
[Table 9 about here]
4 Some Exploratory Regressions
We take the …rst more systematic look at the relationships in the date by study- ing whether three di¤erent norms matter for the return share.6 The …rst is equality. We therefore construct a variable (EQUALITY) for the share of the resources at Borrowers’disposal that should be given to lenders to create perfect equality within each Borrower-Lender relationship. The second norm we look at is need. Given the huge income disparities between Tanzania and Norway, we de…ne the variable NEED to take on the value of 1 if the Borrower is from Norway and the Lender is from Tanzania and zero otherwise. The third norm
5It should be noted, though, that enforcing contracts is costly even when formal institutions for this purpose exist. Hence, even with a well-functioning judicial system enforcement should be expected to be less than 100%.
6In the background, we have in mind a model where individuals dislike deviating from the norms they hold. See e.g. Cappelen et al. (2008) for an example. We plan to add a variant of such a model in future versions.
is related to the obligation to full…l the terms of contracts one has freely en- tered into. To see whether such an invisible handshake matters, we construct a variable (HANDSHAKE) for the contractual obligations as a share of the total resources at Borrowers’disposal.
Table 10 shows that all these norms seem to matter in our experimental data. The three variables are statistically signi…cant at conventional levels.
However, it is clear that in terms of economic signi…cance, considerations of equality and contractual obligations trump needs. Evaluated at the average of the data used for the regressions the elasticities of the return share with respect to EQUALITY, NEED, and HANDSHAKE are equal to 0.436, 0.058, and 0.591, respectively. Thus, while a 1% increase in NEED only generates a 6% increase in the return share, the same change in EQUALITY leads to a 44% increase and a 1% rise in HANDSHAKE produces an increase of 60%.
[Table 10 about here]
Table 11-14 demonstrate that these results are generally very robust to con- trolling for the type of contract, the realised investment multiplier, and the nationalities of Lenders and Borrowers. The only exception is that the size of the multiplier (NATURE), while not signi…cant on its own, greatly increses the size of the coe¢ cient on EQUALITY and its statistical signi…cance while the coe¢ cient on HANDSHAKE decreases and this variable becomes insigni…cant at conventional levels.
[Table 11 about here]
[Table 12 about here]
[Table 13 about here]
[Table 14 about here]
We perform some further robustness tests by including additional variables in the regressions. These are designed to capture potentially more complex e¤ects. First of all, as surplus sharing might be seen as a more fair contract type than the …xed interest variety, Borrowers might be more willing to full…l their obligations in the …rst case. We thus create an interaction variable called FAIRDEAL that is the product of contract type and HANDSHAKE. As …xed interest contracts are coded 1 in our data set and surplus sharing 2, we expect the coe¢ cient of this variable to be positive.
Secondly, the extent to which the full…lment of contractual obligations is a norm might vary between countries. In particular, Norwegians might be more used to both the formal enforcement of contracts by well-functioning juridical institutions as well as to informal loans being repaid as borrowers rarely need to claim "excusable default". The negative correlation of the nationality of Borrowers (country2) and HANDSHAKE shown in Table 9 suggests that such an e¤ect might exist, as the nationality of both Lenders and Borrowers are coded 1 for Norwegians and 2 for Tanzanians. By interacting the nationality of the borrower with the size of the contractual obligations we will capture this e¤ect if it is present in a multivariate context as well. If so, given our coding system the coe¢ cient of the variable COUNTRYNORM should be negative.
Thirdly, the return share might be a¤ected in cases where the nationalities of Borrowers and Lenders di¤er. If people are not moral cosmopolitans, they
care more about their compatriots than foreigners. We thus create a variable that is the absolute value of the di¤erence in the nationalities of the parties to a contract. As the coding of nationalities are the same for Borrowers and Lenders, COMPATRIOTS should be negatively correlated with the return share if people are not moral cosmopolitans. If the variable is not signi…cant, it is an indication that they are and thus do not concern themselves with the nationality of their contractual counterparts.
Fourthly, the size of the loan o¤er might be construed by Borrowers as a sign of trust. In fact, the share of observations for which nothing is returned is more than twice as high for loan o¤ers of only 10 US dollars compared to higher loans.7 The variable TRUST is de…ned as the size of the loan as a share of Lenders’endowment.
Fiftly, while there is extensive evidence that people hold moral norms that induce them to deviate from the behaviour that pure self-interest dictates, one might suspect that the strength of their convictions could be contingent on circumstance. For example, if they experience "bad luck" people might convince themselves, with some justi…cation, that there is less reason to follow their moral rules strictly. To check whether norms are contingent on the degree of luck people have, we therefore create three interaction variables by multiplying NATURE with EQUALITY, NEED, and HANDSHAKE, respectively. As we have already seen, Borrowers seem to be able to rationalise keeping a greater share for themselves when the investment multiplier is three. Thus, a priori it is not clear what sign one should expect these variables to have.
The results are contained in Tables 14-20. The overall impression is once again one of our simple model being remarkably robust. It is also noteworthy that Occam’s Razor is doing well here: none of the additional variables, plausible though they are, are signi…cant.
[Table 15 about here]
[Table 16 about here]
[Table 17 about here]
[Table 18 about here]
[Table 19 about here]
[Table 20 about here]
[Table 21 about here]
The few results of note are the following. The coe¢ cients on FAIRDEAL, COUNTRYNORM, COMPATRIOTS, and TRUST, for which we had clear prior opinions, all had the expected sign. Interestingly, the three variables for norm contingency, for which we had no clear priors with respect to sign are all neg- ative. While they, like the other additional variables are not signi…cant, they hint at people not using bad luck as an excuse to abandon their moral norms (so these are not "luxury goods"); if anything, slacking on morality happens when luck smiles to you.
The coe¢ cient on EQUALITY is increased in terms of both size and signif- icance by the inclusion of the variables capturing the possibilities of contingent
7Recall too, that the acceptance rate was exceptionally low for these loan o¤er.
norms. When the variable TRUST is included, EQUALITY loses its signi…- cance.
The coe¢ cient on NEED increases quite a bit when COMPATRIOTS is in- cluded. This is also the case when we control for whether this norm is contingent on having luck, but then NEED loses its statistical signi…cance.
The coe¢ cient on HANDSHAKE is increased quite some by the inclusion of COUNTRYNORMS. While the latter variable is not signi…cant this at least provides a weak indication that taking into account heterogeneity across coun- tries in terms of the degree to which individuals take contractual obligations seriously only strengthens the …nding that contracts matter. The variables for contingent norms all reduce the coe¢ cient on HANDSHAKE, but it retains its statistical signi…cance in all cases.
5 Preliminary Conclusions
In sum, our …rst analysis of a simple model of norm-guided behaviour suggests that equality, need, and contracts all matter. Our attempts at qualifying their impact by investigating interaction e¤ects or more complex normative consid- erations that seem plausible too are mostly in vain. However, the attraction of equality and the norm of ful…lling the obligations in contracts one has freely entered into clearly dominate the concern for the needs of your contractual part- ners. Moreover, it is interesting to see that what seems to matter is the act of entering into contracts, not the form of the obligations incurred.
References
[1] Camerer, C.F. (2003).Behavioral Game Theory: Experiments in Strategic Interaction. Princeton University Press.
[2] Cappelen, A. et al. (2008). Rich Meets Poor - an International Fairness Experiment. Mimeo.
[3] Easterly, W. and R. Levine (2003): “Tropics, Germs, and Crops: how Endowments In‡uence Economic Development.”Journal of Monetary Eco- nomics 50: 3-39.
[4] Greif, A. (2006): Institutions and the Path to the Modern Economy. Lessons from Medieval Trade. Cambridge University Press.
Table 1: Number of Participants per Session
Session Norway Tanzania Total 20/2: 0900 20 20 40
20/2: 1100 20 21 41 20/2: 1300 20 21 41 22/2: 0900 20 22 42 22/2: 1100 18 17 35 22/2: 1300 20 22 42
Table 2: Demographic Characteristics of Participants by Country
Variable Mean Std. Dev. Min Max Norway: 118 Participants
Gender 1.559322 0.4985856 1 (Male) 2 (Female) Age 23.14407 3.62865 19 41 Year of
Study
2.415254 1.134737 1 4 Tanzania: 123 participants
Gender 1.479675 0.50163 1 (Male) 2 (Female) Age 24.21138 2.162467 20 37 Year of
Study
2.544715 0.6805498 1 4
Table 3: Percent of Participants with Role of Lender by Gender and Nationality
Norway Tanzania Total
Male 44.23 53.13 49.14
Female 54.55 49.15 52.00
Total 50.00 51.22
Table 4: Histogram over Lenders’ Choice
Frequency Percentage Cumulative
0 31 6.57 6.57
10 73 15.47 22.03
20 71 15.04 37.08
30 104 22.03 59.11
40 92 19.49 78.60
50 101 21.40 100.00
Total 472 100.00
Figure 1
Table 5: Was Loans Accepted?
Country of lender Not relevant Yes No Norway 25 (10.68) 147 (62.82) 62 (26.50) Tanzania 6 (2.52) 163 (68.49) 69 (28.99) Total 31 (6.57) 310 (65.68) 131 (27.75) Note: % in parenthesis
Table 6: Acceptance Rate by Nationality of Lender and Borrower
Lender / Borrower Norway Tanzania
Norway 82.70 58.10
Tanzania 76.99 63.87
Note: Excluding situations in which there was no choice to be made
Table 7: Acceptance Rates for Loans by Type of Contract
Country of Borrower Fixed Rate Surplus Sharing
Norway 80.55 78.90
Tanzania 63.39 58.93
Note: Excluding situations in which there was no choice to be made
Table 8: Acceptance Rates for Loans by Size of Loan Offer
Loan / Country of Borrower Norway Tanzania
10 67.57 41.67
20 88.24 56.76
30 82.14 52.08
40 79.55 66.67
50 80.43 80.00
Table 9
returnshare equality need handshake contract nature country1 country2 returnshare 1.0000
equality 0.2995 (0.0000)
1.0000
need 0.0903 (0.1124)
-0.0212 (0.6467)
1.0000
handshake 0.3378 (0.0000)
0.5097 (0.0000)
0.0604 (0.1901)
1.0000
contract 0.0459 (0.4204)
0.0653 (0.1568)
0.0000 (1.0000)
-0.0269 (0.5601)
1.0000
nature -0.1079 (0.0577)
0.1532 (0.0008)
-0.0935 (0.0424)
-0.0611 (0.1848)
0.0231 (0.6162)
1.0000
country1 0.0473 (0.4068)
0.0271 (0.5569)
0.5725 (0.0000)
0.0210 (0.6493)
-0.0000 (1.0000)
-0.1411 (0.0021)
1.0000
country2 -0.0510 (0.3713)
0.0660 (0.1523)
-0.5774 (0.0000)
-0.1222 (0.0079)
0.0000 (1.0000)
0.0334 (0.4690)
0.0085 (0.8543)
1.0000
Notes: p-values in parenthesis. Country1 records nationality of Lender (1=Norway, 2=Tanzania) and country2 nationality of Borrower (with the same coding of countries).
Table 10
Source | SS df MS Number of obs = 310 ---+--- F( 3, 306) = 16.23 Model | 1.35924903 3 .453083012 Prob > F = 0.0000 Residual | 8.54110656 306 .027912113 R-squared = 0.1373 ---+--- Adj R-squared = 0.1288 Total | 9.90035559 309 .032039986 Root MSE = .16707
--- returnshare | Coef. Std. Err. t P>|t| [95% Conf. Interval]
---+--- equality | .2691873 .1429479 1.88 0.061 -.0120981 .5504726 need | .0517446 .0214105 2.42 0.016 .0096142 .093875 handshake | .3542431 .1114089 3.18 0.002 .1350186 .5734676 _cons | -.0212784 .0428248 -0.50 0.620 -.1055467 .06299 ---
Table 11
Source | SS df MS Number of obs = 310 ---+--- F( 4, 305) = 12.25 Model | 1.37039671 4 .342599178 Prob > F = 0.0000 Residual | 8.52995888 305 .027967078 R-squared = 0.1384 ---+--- Adj R-squared = 0.1271 Total | 9.90035559 309 .032039986 Root MSE = .16723
--- returnshare | Coef. Std. Err. t P>|t| [95% Conf. Interval]
---+--- equality | .2538303 .1451414 1.75 0.081 -.0317749 .5394355 need | .0515392 .021434 2.40 0.017 .009362 .0937165 handshake | .3632675 .1124309 3.23 0.001 .1420291 .5845058 contract | .0121687 .0192742 0.63 0.528 -.0257585 .050096 _cons | -.0247418 .0432165 -0.57 0.567 -.1097821 .0602984 ---
Table 12
Source | SS df MS Number of obs = 310 ---+--- F( 4, 305) = 12.63 Model | 1.4071844 4 .3517961 Prob > F = 0.0000 Residual | 8.49317119 305 .027846463 R-squared = 0.1421 ---+--- Adj R-squared = 0.1309 Total | 9.90035559 309 .032039986 Root MSE = .16687
--- returnshare | Coef. Std. Err. t P>|t| [95% Conf. Interval]
---+--- equality | .4154443 .1811422 2.29 0.023 .0589976 .7718909 need | .0506103 .0214027 2.36 0.019 .0084946 .0927261 handshake | .2308236 .1457103 1.58 0.114 -.0559011 .5175483 nature | -.0201431 .0153527 -1.31 0.190 -.0503537 .0100674 _cons | .011847 .0496697 0.24 0.812 -.0858917 .1095857 ---
Table 13
Source | SS df MS Number of obs = 310 ---+--- F( 4, 305) = 12.16 Model | 1.36208464 4 .340521161 Prob > F = 0.0000 Residual | 8.53827095 305 .027994331 R-squared = 0.1376 ---+--- Adj R-squared = 0.1263 Total | 9.90035559 309 .032039986 Root MSE = .16732
--- returnshare | Coef. Std. Err. t P>|t| [95% Conf. Interval]
---+--- equality | .2713699 .1433225 1.89 0.059 -.0106561 .553396 need | .0468478 .0263911 1.78 0.077 -.0050838 .0987794 handshake | .3544561 .1115749 3.18 0.002 .1349022 .5740101 country1 | .0075445 .0237053 0.32 0.751 -.039102 .0541911 _cons | -.0323839 .0552898 -0.59 0.559 -.1411816 .0764138 ---
Table 14
Source | SS df MS Number of obs = 310 ---+--- F( 4, 305) = 12.15 Model | 1.36090606 4 .340226516 Prob > F = 0.0000 Residual | 8.53944953 305 .027998195 R-squared = 0.1375 ---+--- Adj R-squared = 0.1261 Total | 9.90035559 309 .032039986 Root MSE = .16733
--- returnshare | Coef. Std. Err. t P>|t| [95% Conf. Interval]
---+--- equality | .2706091 .1432875 1.89 0.060 -.0113479 .5525662 need | .0483405 .0256052 1.89 0.060 -.0020446 .0987256 handshake | .3538848 .1115903 3.17 0.002 .1343005 .5734691 country2 | -.0056064 .0230454 -0.24 0.808 -.0509546 .0397418 _cons | -.0126634 .0556205 -0.23 0.820 -.1221119 .096785 ---
Table 15
Source | SS df MS Number of obs = 310 ---+--- F( 4, 305) = 12.19 Model | 1.36481655 4 .341204136 Prob > F = 0.0000 Residual | 8.53553905 305 .027985374 R-squared = 0.1379 ---+--- Adj R-squared = 0.1265 Total | 9.90035559 309 .032039986 Root MSE = .16729
--- returnshare | Coef. Std. Err. t P>|t| [95% Conf. Interval]
---+--- equality | .2509977 .1488316 1.69 0.093 -.0418689 .5438644 need | .0517872 .0214388 2.42 0.016 .0096006 .0939738 handshake | .3595677 .112192 3.20 0.001 .1387995 .580336 fairdeal | .0206164 .0462219 0.45 0.656 -.0703377 .1115705 _cons | -.0203524 .0429312 -0.47 0.636 -.1048311 .0641264 ---
Table 16
Source | SS df MS Number of obs = 310 ---+--- F( 4, 305) = 12.47 Model | 1.391505 4 .347876251 Prob > F = 0.0000 Residual | 8.50885059 305 .027897871 R-squared = 0.1406 ---+--- Adj R-squared = 0.1293 Total | 9.90035559 309 .032039986 Root MSE = .16703
--- returnshare | Coef. Std. Err. t P>|t| [95% Conf. Interval]
---+--- equality | .2742566 .1429892 1.92 0.056 -.0071136 .5556269 need | .0379245 .0249673 1.52 0.130 -.0112055 .0870544 handshake | .4331637 .1333887 3.25 0.001 .1706851 .6956422 countrynorm | -.055564 .0516742 -1.08 0.283 -.1572471 .0461191 _cons | -.018747 .0428785 -0.44 0.662 -.1031221 .0656282 ---
Table 17
Source | SS df MS Number of obs = 310 ---+--- F( 4, 305) = 12.24 Model | 1.36959317 4 .342398293 Prob > F = 0.0000 Residual | 8.53076242 305 .027969713 R-squared = 0.1383 ---+--- Adj R-squared = 0.1270 Total | 9.90035559 309 .032039986 Root MSE = .16724
--- returnshare | Coef. Std. Err. t P>|t| [95% Conf. Interval]
---+--- equality | .2775518 .1437549 1.93 0.054 -.0053251 .5604287 need | .0632034 .0285375 2.21 0.028 .0070481 .1193587 handshake | .3536937 .1115275 3.17 0.002 .134233 .5731543 compatriots | -.0154066 .025334 -0.61 0.544 -.0652582 .034445 _cons | -.0202827 .0429002 -0.47 0.637 -.1047005 .0641352 ---
Table 18
Source | SS df MS Number of obs = 310 ---+--- F( 4, 305) = 12.18 Model | 1.36330473 4 .340826182 Prob > F = 0.0000 Residual | 8.53705087 305 .027990331 R-squared = 0.1377 ---+--- Adj R-squared = 0.1264 Total | 9.90035559 309 .032039986 Root MSE = .1673
--- returnshare | Coef. Std. Err. t P>|t| [95% Conf. Interval]
---+--- equality | .1589206 .3231174 0.49 0.623 -.4769009 .7947421 need | .0518725 .0214431 2.42 0.016 .0096774 .0940676 handshake | .3385075 .1189773 2.85 0.005 .1043872 .5726278 trust | .0465186 .1222076 0.38 0.704 -.1939582 .2869954 _cons | -.0011859 .0680094 -0.02 0.986 -.1350129 .132641 ---
Table 19
Source | SS df MS Number of obs = 310 ---+--- F( 4, 305) = 12.46 Model | 1.39037791 4 .347594477 Prob > F = 0.0000 Residual | 8.50997768 305 .027901566 R-squared = 0.1404 ---+--- Adj R-squared = 0.1292 Total | 9.90035559 309 .032039986 Root MSE = .16704
--- returnshare | Coef. Std. Err. t P>|t| [95% Conf. Interval]
---+--- equality | .4566818 .2278946 2.00 0.046 .0082371 .9051266 need | .0510575 .0214163 2.38 0.018 .0089151 .0931999 handshake | .2554715 .145436 1.76 0.080 -.0307135 .5416565 expostnormE | -.0389719 .0368965 -1.06 0.292 -.1115758 .0336319 _cons | -.0234338 .0428653 -0.55 0.585 -.107783 .0609153 ---
Table 20
Source | SS df MS Number of obs = 310 ---+--- F( 4, 305) = 12.25 Model | 1.3701909 4 .342547725 Prob > F = 0.0000 Residual | 8.53016469 305 .027967753 R-squared = 0.1384 ---+--- Adj R-squared = 0.1271 Total | 9.90035559 309 .032039986 Root MSE = .16724
--- returnshare | Coef. Std. Err. t P>|t| [95% Conf. Interval]
---+--- equality | .3045818 .1538732 1.98 0.049 .0017943 .6073692 need | .0813473 .051954 1.57 0.118 -.0208863 .1835809 handshake | .3263629 .1200979 2.72 0.007 .0900375 .5626882 expostnormN | -.0153694 .0245719 -0.63 0.532 -.0637213 .0329826 _cons | -.024153 .0431131 -0.56 0.576 -.1089898 .0606837 ---
Table 21
Source | SS df MS Number of obs = 310 ---+--- F( 4, 305) = 12.38 Model | 1.38330554 4 .345826384 Prob > F = 0.0000 Residual | 8.51705006 305 .027924754 R-squared = 0.1397 ---+--- Adj R-squared = 0.1284 Total | 9.90035559 309 .032039986 Root MSE = .16711
--- returnshare | Coef. Std. Err. t P>|t| [95% Conf. Interval]
---+--- equality | .3814665 .1872889 2.04 0.043 .0129246 .7500084 need | .0510334 .021429 2.38 0.018 .0088659 .0932008 handshake | .3197495 .1174678 2.72 0.007 .0885995 .5508995 expostnormH | -.0350102 .0377202 -0.93 0.354 -.1092349 .0392144 _cons | -.0238366 .0429231 -0.56 0.579 -.1082994 .0606262 ---