Entrepreneurship and Liquidity Constraints in Rural Thailand

Report on Research in Progress

Anna Paulson

Northwestern University

Robert Townsend

University of Chicago

April 1999

Preliminary

Please do not cite or quote without permission of the authors.

From the beginning of the 1980’s until the mid 1990’s, the Thai economy grew very rapidly and the financial sector grew even faster than the country as a whole. During this period of rapid growth and financial deepening, there were also substantial increases in inequality. With the devaluation of the baht in 1997, Thailand became the epicenter of the Asian financial crisis. The country itself experienced a deep recession. Now it appears that Thailand may have turned the corner and may be on the road to recovery. In order to understand these phenomenon, we need both theory and data. In this paper, we take a step in that direction, focusing on the period of growth and increasing inequality and uneven financial deepening. We make use of data that we collected in Thailand in the spring of 1997, prior to the devaluation. The survey instruments that were used to collect the data were designed with various strands of theoretical and empirical literatures very much in mind.

We consider three models, all of which can generate economic growth and increasing and then decreasing inequality. All of the models emphasize the role of occupational choice in developing these dynamics. The first model that we consider comes from Lloyd-Ellis and Bernhardt (1996). This is a dynamic growth model that features increasing and then decreasing inequality as the economy develops. In this model, entrepreneurial talent is important in determining the fixed cost of becoming an entrepreneur but has no further effect on productivity. Agents must finance entrepreneurial investment entirely out of their inherited wealth. There is no credit market in the model, although one can imagine the impact of an innovation that allows for financial intermediation as in Townsend, et. al. (1999). Not everyone in this model will become an entrepreneur and some (but not all) entrepreneurs will be constrained.

The second model we consider is developed in Evans and Jovanovic (1989) and featured in Holtz-Eakin, Joulfian and Rosen (1994). This static model emphasizes the role of entrepreneurial talent in determining the optimal level of business investment and allows for a limited credit market. Borrowing is exogenously limited to be at most some multiple of household wealth. This borrowing limit could be made endogenous by considering default or limited commitment as in Banerjee and Newman (1993), where a higher probability of being caught upon default would allow for a greater multiple of wealth to be borrowed. As in Banerjee and Newman as well, these micro underpinnings could be used to model growth and the distribution of income.

The third model we consider is Lehnert (1999) which builds on Piketty (1997) and Aghion and Bolton (1996). These are private information models where agents make decisions about how much capital to invest and how hard to work. Effort is unobservable, however. In these models, credit is limited endogenously as the solution to an information constrained contracting problem, rather than directly through assumptions about collateral requirements. However, some profitable projects may not be funded in this economy and the amount of credit will vary with wealth. Credit is not necessarily an increasing function of wealth or even monotonically related to wealth. These models also deliver relationships between growth and inequality.

These types of models appear to be reasonable candidates for explaining the story of growth with increasing inequality and uneven financial deepening. Yet, as we emphasize below, the models have different implications for occupation choice (entrepreneurship v. wage labor); the relationship between financing constraints, wealth and entrepreneurial talent; and the relationship between investment, net savings and wealth. In this paper, we report our preliminary and on-going efforts to see which implications of the models are borne out in the data. Our eventual goal is either to modify and nest the models so than their overall fit can be compared, or to use non-nested techniques to compare them. We reiterate, however, that the step that we take here, comparing and contrasting salient patterns in the data with the implications of the models, is an illuminating step on the way toward this longer-run goal.

The rest of the paper is organized as follows. In the next section we provide background information on trends in Thai income growth, financial development and income distribution over the last two decades. This section also highlights the key geographical differences and describes the important financial institutions. The third and fourth sections describe the larger project and describe the household and small business data that is used in the analysis. In the fifth section, we discuss the three models and their implications in detail. Empirical tests of these implications are presented and discussed in Section 6. This section also provides tentative conclusions and plans for continuing research.

2. General Thai Background

From the beginning of the 1980’s until the mid 1990’s, the Thai economy grew very rapidly. From 1981 to 1995, average income in Thailand grew at an average real rate of 8% per year. While incomes have grown for virtually all segments of the population during this period of economic expansion, the income of the richest group has grown the most, leading to substantially increased inequality. From 1981 to 1992, Thailand’s gini coefficient increased 122% going from 0.44 to 0.54 (Jeong, 1998). Inequality appears to have decreased after 1992, with the 1996 gini coefficient falling to 0.50.

This period of rapid economic development has been accompanied by an even more rapid deepening of the financial sector, at least at the macro level. In 1990 the ratio of M2 to GDP in Thailand was the same as the ratio in the U.S. at 70%. By 1996, the Thai ratio had increased to 81%. While much of the financial sector development in Thailand was focused in urban areas in and around Bangkok, financial services to rural areas have also increased. From 1988 to 1994, the percentage of villages where at least one person borrowed from a commercial bank increased by 28% (calculations from the Thai Community Development Department (CDD) Village surveys). The fraction of villages where someone borrowed from the Bank for Agriculture and Agricultural Cooperatives (BAAC) increased by 11% and the fraction of villages where someone borrowed from a village Production Credit Group (PCG) increased by 33%.

Despite these figures, the use of financial services in rural areas differs in important ways from the country as a whole. For example, commercial banks play a large role in the overall economy, accounting for 67% of total deposits and 67% of total credit (Bank of Thailand report). But according to the CDD data, in 1994, commercial banks had one or more customers in only 35% of villages. There are also regulatory and political restrictions that may have limited intermediation through commercial banks. In the rural areas we analyze, commercial banks hold 56% of deposits and account for only 19% of total funds lent out (Seiler and Townsend, 1999 and Kaboski and Townsend, 1999). Only 4% of the households we surveyed currently have debts to commercial banks.

Other financial institutions that are important at a national level are even more absent from rural areas. For example, finance companies accounted for 17% of total Thai savings and 21% of total credit. None of the households that we analyze are customers of finance companies. A model with no credit would not seem far from reality; at least in so far as commercial banks and finance companies are concerned.

In contrast to commercial banks and finance companies, the BAAC plays a very large role in rural areas but plays only a small role in the overall financial picture. Nationally, the BAAC accounts for 1.5% of savings and 2.3% of credit. But according to the CDD data, the BAAC has customers in 86% of villages in 1994. In our study area, 15% of total savings are on deposit at the BAAC and 29% of total credit is provided by this institution. The BAAC lends up to 60,000 baht on social collateral through joint liability groups. For loans above that amount, physical collateral in the form of land or other fixed assets is required, much like commercial banks. This pattern is reminiscent of the models that allow credit to equal a fixed multiple of wealth. Certainly loan repayment is enforced, with collateral confiscation. Legal limitations which, until recently, restricted BAAC lending to agriculture and agriculturally related enterprises are also suggestive of limited credit models. On the other hand the BAAC can be a fairly flexible institution. For example, we find businesses in our survey that are financed through BAAC credit and the BAAC has a risk-contingency system under which adverse events are verified and flexible repayment plans are offered (see Townsend and Yaron, forthcoming).

Village PCGs are unmeasured in the national level flow of funds accounts. According to the CDD data they were providing credit in 33% of villages in 1994. They provide only a small fraction of savings and credit in the study area: 3.3% of savings and 1.2% of credit. PCGs can be viewed as part of a set of government programs to target the poor, to redress inequality issues. Despite being responsible for only a small percentage of total savings and credit, PCGs are very important in some villages and can in principle be viewed as innovations in the availability financial intermediation.

Informal savings and credit are conjectured but virtually unmeasured at the national level. Surveys by the Thai Development Research Institute find that the geographic scope of the informal sector has declined substantially over the past ten years. Yet informal sources of savings and credit are as important as institutional sources of deposit and credit services in the study area. Many households, especially in the poorer northeastern region, have savings in the form of rice. Seventy-one percent of households have rice in storage in the Northeast and it makes up a large part of household savings. On average, savings in the form of goods (almost exclusively rice) account for 56% of total household savings in the Northeast. While the figures are much smaller in the Central region, savings in the form of rice still make up a substantial fraction of total household savings. In Lopburi 10% of total savings is in the form of goods, on average. In Chachoengsao, the figure is 15%. Informal sources of credit are also very important: relatives account for 14% of total credit in the study area. Shop owners and moneylenders account for another 6% and 7% of total credit respectively. Purchasers of output account for 3% of total credit and neighbors provide an additional 2%. Even in Lopburi, which is relatively developed and close to Bangkok, 43% of total credit comes from informal sources.

In general, informal loans are smaller, less likely to require collateral, have higher interest rates and are of shorter duration, compared with formal sector loans. Informal loans tend to have more flexible repayment arrangements, however. For example, interest rates and the repayment date may not be specified in advance and penalties for non-payment seem less severe. These features are reminiscent of private information models.

Aggregating over all sources, 32% of our survey households report no loans over the past 12 months. About half of the households with out-standing loans have borrowed from only one source. The borrowing to income ratio is 33% overall, but is as high as 300% for households in the lower third of the income distribution. The interest rate moves inversely with income. The collateral to loan ratio is 15:1 for loans that require collateral, ranging from 8:1 for loans from neighbors to 14:1 for the BAAC and 26:1 for relatives and 31:1 for loans from commercial banks.

Our theoretical and empirical focus on business start-ups and the financial sector is motivated in part by how occupational change and financial development have shaped Thailand’s experience of economic growth and the accompanying increase in inequality. In his analysis of the 1976 – 1996 Thai Socio Economic Surveys, Jeong (1998) finds that 39% of Thailand’s growth can be attributed to three factors: shifts in the occupational distribution, changes in education and in the expansion of financial intermediation. These same factors account for 53% of the increase in inequality 1976 to 1996. Occupational shifts have been particularly important in rural areas.

The importance of these factors is also confirmed by the responses of the households that we surveyed. One-third of households report that they would like to change occupations. Of the households who would like to change occupations, most would like to open a business. Households who run some kind of small business make up about one-fifth of the sample. The average income of these households is three times higher than that of non-business owners. Fifty-four percent of households who do have small businesses report that their business would be more profitable if they could expand it. When they are asked why they do not undertake this profitable opportunity, 56% of households report that they do not have enough money to do so.

3. Project Background and the Survey Data

The data that we analyze in this paper are the product of a large on-going socio-economic/institutional study in Thailand that is funded by the National Institute of Health and the National Science Foundation in the U.S. through the University of Chicago/NORC. The initial survey of households, village financial institutions and village key informants was completed in May of 1997 and covers regions both on the doorstep of Bangkok as well as in the relatively poor northeast. The data provide a wealth of pre-financial crisis socio-economic and financial data on 2880 households, 606 small businesses, 192 villages, 161 local financial institutions, 262 borrowing groups of the BAAC and soil samples from 1880 agricultural plots. We focus on the household survey data in this paper.

The Ford Foundation funded a follow-up survey of 33% of the original sample households and institutions. This survey was completed in the spring of 1998 and provides information on the impact of the crisis. A third follow-up survey is being administered in the spring of 1999. In addition to these data, a smaller and more intensive set of data gathering activities began in sixteen of the original sample villages in July of 1998. Each of the 45 sample households in the selected villages will be surveyed once a month for two years. The Association of Production Credit Groups in Thailand is responsible for implementing each of these surveys. All of these data collection activities are aided by the cooperation of the BAAC and the Community Development Department of the Thai Ministry of the Interior.

The data we analyze cover four provinces in Thailand. Two of the provinces, Lopburi and Chachoengsao are in the Central region and are relatively close to Bangkok. Chachoengsao borders the Bangkok Metropolitan Area and forms part of the industrial corridor that extends to Thailand’s eastern seaboard. The other two provinces, Buriram and Sisaket are much further from Bangkok and are located in the relatively poor northeastern region. Sisaket is one of the poorest provinces in the country. The contrast between the survey areas is deliberate and has obvious advantages.

The choice of the provinces was also driven by the fact that at least one amphoe (county) in each province was included in five successive waves of the Thai Socio Economic Survey (SES) which is conducted by the Thailand’s National Statistical Office. These data were available for 1976, 1981, 1986, 1988 and 1990 at the time the provinces were chosen and they provide baseline historical information for the study area.

In each of the four provinces, a stratified random sample of twelve tambons (subset of an amphoe or county) was chosen. The stratification ensured an ecologically balanced sample that included two "forested" tambons. Within each sample tambon, four villages were selected at random. Fifteen households were randomly selected from each of the sample villages. Soil samples were collected from randomly chosen plots belonging to the first ten of these households. In each village a key informant was interviewed, usually the village headman or woman. In addition, interviews were conducted with the committee members of each village financial institution. If the sample village included BAAC borrowing groups, they were also interviewed. The sampling strategy will be described in more detail in Binford, Lee and Townsend (1999).

The survey instruments were designed to incorporate the latest advances in economic theory. The theories that guided the design of the questionnaires include models of liquidity constraints in human capital formation, in occupation choice and in input financing; models of growth with costly financial deepening; and models which take into consideration private information and incentives. We proceeded by extrapolating from the theory the true story underlying Thailand’s growth and the accompanying increase in inequality and financial deepening might have been. Or more formally, by writing down and simulating a variety of economic models. This process generated a list of variables that were crucial to assessing the validity of a particular story or economic model. This list of variables formed the basis for designing the questionnaires and the field research.

Because these data provide rich and detailed information about both the firm and the entrepreneurial household as well as information on financial intermediaries, they are particular well designed for studying theories of entrepreneurship and the financial system. Economic theory emphasizes that both firm and household characteristics are important in determining the supply and the demand for credit. In many studies the available data force a focus on either the firm or the entrepreneur, but do not allow for both to be treated with equal thoroughness. For example, Evans and Jovanovic (1989) use data from the National Longitudinal Study of Young Men, which has detailed information on the self-employed, but very sparse information about the businesses they run. In their studies of banking relationships and access to credit in small businesses, Petersen and Rajan (1994 and 1995) analyze data collected by the Small Business Administration (SBA). The SBA data provide a wealth of detail about the firm, but almost no information about the entrepreneur. Holtz-Eakin, Joulfian and Rosen (1994) use data from U.S. individual tax returns. These data provide detailed information about inheritances and some information about both the entrepreneur and the firm, however they do not include important firm and household variables, the nature of the business and the education of the entrepreneur, for example.

The data that we analyze here provide detailed information about the household as well as about the firm. The data include information on household composition and education, income, expenditures, characteristics of land holdings, household and agricultural assets. Information about how and when assets and inheritances were acquired is also recorded. In addition, the data also include information about outstanding debts, loans that the household has made, and information about savings. Current and past occupations of all household members are also recorded. The household data also provide information about whether the household is a (or has been) a member or a customer of a number of formal and informal financial institutions.

The firm data include the type of business, the amount and sources of start-up investment as well as the amount and sources of current business investment. The age of the firm, the number of family and non-family workers and current profit data are also available. The retrospective data on wealth, inheritances and interactions with financial institutions helps are efforts to disentangle the effects of running a business from the forces which make it possible for a household to start a business in the first place.

Table 1A provides a summary of household characteristics by region and whether or not the household has a business. Twenty-one percent of the sample households have a business, and these businesses are concentrated in the relatively prosperous Central region where 28% of the households have a business. In the Northeast, only 13% of households have a business. Business owners tend to be a bit younger and substantially more educated than their non-entrepreneurial counterparts. This is especially true in the Northeast, where 31% of the heads of business owning households have more than four years of schooling compared with only 14% of non-business owning heads. In the Central region the difference is less dramatic: 21% of the heads of business owning households have more than four years of schooling compared with 15% of the heads of non-business households. The percentage of business owning households who have less than 4 years of schooling is consistently lower than for non-business households. Median annual income is considerably higher for business owners. In both the Northeast and the Central region, the median income of business households is about twice that of non-business households.

In addition to having higher incomes, business households are also wealthier. Table 1B examines the magnitudes and the sources of real and financial wealth for business and non-business households in each region. In the Northeast, the average wealth of business owners is 15% higher than that of non-business owners. However, median wealth is 54% higher than that of non-business owners. The difference between business and non-business households is even larger in the Central region. Average wealth is 200% higher and median wealth is 161% higher. Sources of wealth also differ across business and non-business households. For example, 76% of business owners have savings in financial institutions, compared with 58% of non-business owners. There are also some interesting differences in the patterns of real asset holdings across business and non-business owners. One thing that stands out is differences in the percentage of households who own titled land. In the Northeast, business owners are more likely to own land with full title compared to non-business owners (52% v. 45%). In the Central region, the opposite is true (29% v. 43%).

In addition to being wealthier, business owners are also more likely to have made loans to other individuals. Twenty percent of business owners in the Northeast are owed money, compared with 11% of non-business owners. The pattern in the Central region is similar: 14% of business households are owed money, compared with 8% of non-business households.

Because households were asked to report when they acquired household and agricultural assets and land, we can get an indication of past wealth as well as current wealth. The past value of real assets is found by depreciating the purchase price of the asset (in 1997 baht) from the time of purchase to what it would have been worth six years ago. We assume that the depreciation rate for all household and agricultural assets is 10% per year. If the household purchased a walking tractor 10 years before the survey for 100,000 baht, we would first convert the purchase price to 1997 baht (using the Thai consumer price index) and then multiply this by (0.90)⁴ to estimate the value of the walking tractor six years prior to the survey. Past values of land are treated differently. Households were asked to report the current value of each plot that they own. In calculating past land values, we assume that there have been no changes in land prices. So if the household has had one plot for ten years and the current value of that plot is 100,000 baht, then six years ago the value of that plot will also be 100,000 baht (in 1997 baht). Because of the way they will be used in the analysis, these measures also do not include the value of land that was inherited between six and ten years ago. These summary statistics are reported in Table 1C.

The indicators of past wealth that we compute are incomplete in (at least) two respects. The first issue is that we only have information on household and agricultural assets that the household still owns. If the household purchased a car 10 years ago and sold it 3 years ago, the value of the car will not be included in past wealth. If the household still owns the car, then it will be included in the measure of past wealth. The second concern is that we do not have any information on past financial assets and liabilities. Since financial assets and liabilities tend to make up a small fraction of current household wealth, leaving them out of our measures of past wealth is probably not too big of a problem.

The measures of past wealth indicate that business households were wealthier than non-business households in the past as well as at the time of the survey, especially in the Central region (see the bottom of Table 1C). Six years ago, business households in the Northeast were 23% wealthier than non-business households were, on average. In the Central region, business households were 163% wealthier on average. The gap between business and non-business households appears to be increasing, particularly in the Central region. Looking at the current counterpart to the past wealth measures, we see that business owners in the Northeast are currently 38% wealthier than non-business households were. In the Central region, the gap between business and non-business households has grown to 207%. Because past wealth is measured imprecisely, these figures should be interpreted with caution. When we compare the difference between all sources of current wealth the same figures would be 15% and 200%.

In addition to retrospective data on wealth, we also have information about when households received inheritances and the value of those inheritances (see Table 1C). The data on inheritances provide measures of exogenous variation in wealth that we use to evaluate the predictions of the theory. In the Northeast, 15% of non-business household received an inheritance between six and ten years ago, compared with 19% of business households. Although fewer households received inheritances in the Central region, the overall pattern is similar: 15% of business households and 11% of non-business households received inheritances between six and ten years ago. The median size of the inheritance does not vary substantially depending on whether the household currently runs a business. In the Northeast, median inheritances for business households (who received inheritances) were 44,000 baht, compared with 56,000 baht for non-business owners. In the Central region, the median business household inheritance was 120,000 and the median non-business household inheritance was 104,000.

Business owners were asked to report how initial asset purchases and other start-up costs were financed. Almost half of the funding for the purchase of assets, 48%, comes from cash and another 26% comes from credit. The reported sources of credit are: relatives (11%), store owners (7%), BAAC (3%), and commercial banks (3%). Gifts accounted for an additional 4% of the funds used to purchase assets. There was no other mention of inheritances.

Approximately 30% of the businesses required start-up funding in addition to what was used to purchase business assets at the time the business was founded. In the Northeast another 52,000 baht was required, on average. In the Central region, 170,000 baht was needed. The important sources of additional funding are very similar across the two regions. The four most important sources are savings (41% of total, on average), the BAAC (15% of total), other (12% of total), and commercial banks (6% of total).

More generally, we can also look at the characteristics of out-standing loans among business and non-business households, ignoring any distinctions between households and their businesses, and also leaving aside issues of selection effects. Most households in the sample have multiple loans and there is a slight tendency for business households to have more outstanding loans than non-business households, 1.3 loans v. 1.2 loans. As a percentage of their portfolio of liabilities, households with businesses tend to borrow more from commercial banks (8%) and the BAAC (36%) than non business owners do, 2% and 34%, respectively. Business owners tend to borrow less from neighbors (4% v. 9%) and moneylenders (8% v. 11%).

Businesses tend to have larger loans from virtually all formal and informal sources. The one exception is loans from village funds. Although business owners have higher incomes, the ratio of loans to income is slightly higher for business owners compared to non-business owners: 0.37 v. 0.31. Overall, interest rates tend to be lower for loans to business households, 14% v. 18%. Although in Lopburi average interest rates for business owners are 21% compared to 18% for non-business households. Despite the overall pattern in interest rates, non-business households have a higher fraction of no-interest loans 19% compared to 15% for business owners. This reflects the tendency of business owners to rely less on the informal sector.

The purpose of the loan also varies depending on whether the household owns a business or not. Fertilizer and pesticide loans are less common for business households (20%) than for non-business households (28%), although these figures reflect the number of loans and not the amount of lending for this purpose. Consumption loans are also less common among business households, 10% v. 20%. Business owners report that loans for the business make up 25% of their total loans. Non-business households also report a few business loans, presumably they are thinking about their agricultural activities.

Business households are more likely to have collateralized loans (76%) compared to non-business households (56%). The percentage of all loans that are collateralized with land is 30% for business owners and 19% for non-business owners. Interestingly, both business and non-business owners have about 25% of their loans secured only by multiple guarantors. Among collateralized loans, the ratio of the value of the collateral to the amount of the loan is a positive multiple of the loan itself. For business owners, collateral value averages 9 times the amount of the loan. For non-business households the ratio is almost twice as high at 17. This pattern holds even for secured loans among relatives. The ratio is 7 for business households and 31 for non-business owners. Since land parcels may not be divisible, this finding may be consistent with business owners receiving more credit. Still aggregating over all financial liabilities and all capital and financial assets, the ratio of the value of assets to loans is still about 41 for non-business households and 25 for business households. In the Northeast, the contrast is even starker: 42 v. 16.

Some of the patterns that we see in the loan data are replicated in information about current and past member patronization of various financial institutions, see Table 1C. We have grouped these institutions into five categories. The first, formal financial institutions, includes commercial banks, finance companies, insurance companies as well as national employee credit unions like the Teachers Credit Union. The second, village institutions and organizations, is made up of PCGs, rice and buffalo banks as well as village poor and elderly funds. The BAAC, the Agricultural Cooperative and local farmer’s groups are included in the third group, agricultural organizations. Moneylenders and Rotating Savings and Credit Associations (ROSCAs) make up the fourth and fifth groups respectively. While most of these organizations provide savings and/or credit services, a few do not. The most prominent type of institution varies across the two regions. In the Northeast, agricultural organizations have the highest participation rate among both business and non-business households. Formal financial institutions have the highest participation rate in the Central region.

In both the Northeast and the Central region, business owners are more likely to be current customers of formal financial institutions, village institutions and organizations and agricultural organizations. Interestingly, about 14% of all households are currently customers of a moneylender and there is essentially no difference in this fraction across regions and across business and non-business households. No households in the Northeast are members of a ROSCA, but ROSCAs do seem to be important in the Central region. Thirteen percent of business households in the Central region are current members of a ROSCA compared with 6% of non-business households.

Households were asked to report when they became a customer or member of each organization, so we are also able to look at past participation in these same organizations. Six years ago, participation in all types of organizations appears to have been much lower and formal financial institutions were less prominent. Agricultural organizations had the highest participation rate for all groups, except business owners in the Central region who were a bit more likely even then to be customers of a formal financial institution. Although past participation rates are lower in general, participation among business owners was still higher that of non-business owners. For example, 11% of non-business owners in the Northeast report being a customer of a formal financial institution six years ago, compared with 18% of northeastern business households. In the Central region, the same figures are 19% and 34%.

Table 1D provides summary information on the businesses that are included in the survey. Of the households with businesses, 89% of households in the Northeast and 79% of households in the Central region have one business. Partnerships and joint business ventures are not important: virtually all (97%) of the businesses are wholly owned by the survey household. Most of these businesses were established in the last five years – 68% of the businesses in the Northeast and 60% of the business in the Central region. In much of the empirical work, we concentrate on these businesses. In each region, another 20% were established between six and ten years ago. Businesses that were established eleven or more years ago make up 11% of the total in the Northeast and 19% of the total in the Central region.

About 70% of the business have only one or two workers and family workers account for account for more than 90% of the regular workforce. However, 6% of the business in the Northeast and 11% of the businesses in the Central region paid someone for work during the past twelve months.

Each respondent with a household business was asked, "If you could increase the size of your business, do you think it would be more profitable?". We call the households who answered yes to this question "constrained". Sixty-four percent of the household businesses in the Northeast and 50% of the households in the Central region answered yes to this question (see the bottom of Table 1D). Those who answered yes were asked to name the main barriers to expanding the business. The most common answer, especially in the Northeast, was "not enough money to expand". Seventy-two percent of the constrained businesses in the Northeast gave this answer, compared with 47% in the Central region. Scarce land and labor were also important reasons for not expanding, particularly in the Central region. Thirteen percent of constrained businesses in the Central region reported that they did not have enough land to expand and 15% reported that they did not have enough labor.

The different types of businesses are described in Table 1E. The businesses are fairly evenly split between those that are related to agriculture (rice threshing, raising shrimp and/or fish, trading crops, livestock and food) and those that are non-agricultural (shops, mechanic and repair shops, services, sewing, restaurants, transport, trade in manufactured goods). Businesses that are related to agricultural are more common in the Central region (55%) than in the Northeast (34%). This is because shrimp and fish raising and livestock activities are concentrated in the Central region. The most common businesses in the Northeast are shops (41%), shrimp and/or fish raising (13%), followed by rice threshing (9%). In the Central region, shrimp and/or fish make up 32% of businesses, followed by shops (18%) and trading livestock or their products (15%). Trade of all types is more prevalent in the Central region, making up 22% of the total number of businesses compared with 12% in the Northeast.

The average initial investment in the household business varies substantially by business type and sometimes by region (see Table 1F). For example, the median initial investment in a shop is 16,000 baht in both the Northeast and the Central region. The median initial investment in livestock is 48,000 baht in the Central region but only 3,000 baht in the Northeast. Some of this variation presumably reflects differences in the type of livestock that are raised in the two regions. Expensive dairy cattle are found almost exclusively in the Central region province of Lopburi while cheaper poultry and pigs are more common in the Northeast. In general, median initial investment in the business is higher in the Central region compared with the Northeast: 38,000 v. 19,000 baht. There is also substantial variation in initial investment within regions. In the Northeast, median investment is 29,000 baht in Buriram compared with only 13,000 baht in Sisaket. In the Central region, Lopburi’s median investment looks like Sisaket’s at only 14,000 baht compared with 48,000 baht in Chachoengsao. This is a somewhat misleading picture of Lopburi, however. At the 75^th percentile, the initial business investment is 90,000 baht in Lopburi compared with 50,000 baht in Sisaket, 78,000 baht in Buriram and 143,000 baht in Chachoengsao. Businesses are also not evenly spread within the two regions, especially in the Central region. Fifty-five percent of the businesses in the Northeast are in Buriram and 64% of the businesses in the Central region are in Chachoengsao.

In this section, we describe the three models whose empirical implications we are interested in evaluating. We also describe and compare the empirical implications of each of the models. The model descriptions below concentrate on describing the key features that are necessary for developing the implications that we evaluate in the empirical section. All of the papers we discuss include much richer descriptions of the model economies than we do justice to here. In on-going work, Townsend et. al. (1999) use numerical simulations of the Lloyd – Ellis and Bernhardt model to explain Thailand’s growth and increasing inequality prior to the crisis and to conduct some welfare experiments. In a related paper, Jeong and Townsend (1999) examine how well this model can explain changes income growth rates and changes in the distribution of income in Thailand. Our emphasis in this paper, however, is in describing and evaluating the micro economic implications of the models. This emphasis on empirical work guides the level of detail we provide in describing the models.

In the model descriptions, the same symbols are used to label variables that are common across models. For example, the variable A always stands for wealth, regardless of the model under consideration, k* always denotes the optimal amount of capital to invest when households are unconstrained and the parameter q * stands for entrepreneurial talent.

This is a dynamic model of occupational choice where households decide whether to work in the wage sector or to become an entrepreneur. Becoming an entrepreneur requires paying a fixed start-up cost. The maximum amount of capital that can be invested in entrepreneurial projects (including the start-up cost) is limited by the amount of wealth the household has inherited. There is no credit market.

Agents in this model start their lives with wealth, A, that they have inherited from their parents. They are also endowed with some amount of entrepreneurial talent that determines the fixed cost of starting a business. More talented individuals have lower fixed costs, q . If agents decide to work in the wage sector, they will receive the prevailing equilibrium wage, w. Agents take the wage as given, although in the model it will change through time as relative numbers of wage workers and entrepreneurs change. Agents can also "save" their inheritance until the end of the period. No interest is paid on this type of savings. If they decide to become entrepreneurs, their earnings are determined by maximizing the following profit function:

where k is the amount of capital invested and l is the amount of labor. The amount of capital that is chosen must be less than inherited wealth net of the fixed cost of establishing the business:

0 £ k £ A - q

The amount of capital that is invested in the business will depend on whether or not this constraint is binding. Let g be the Lagrange multiplier on the constraint. First order conditions from maximizing profits will be:

f_k(k, l) – 1 - g = 0

f_l(k, l) – w = 0

A - q - k = 0

When the constraint is binding, the amount of capital invested will be equal to A - q and will clearly depend on wealth (A) and the start-up cost (q ). It will be independent of the equilibrium wage (w). When the constraint is binding, any increase in wealth is invested in the business.

When the constraint does not bind and g is equal to zero, the amount of capital invested will be independent of start-up costs and the amount of wealth that has been inherited. In the unconstrained case, the amount of capital invested will depend only on the equilibrium wage, w.

The implications of this model for the choice between being a wage worker, starting a business and being constrained or unconstrained are summarized in Figure 1 for a given equilibrium wage, w. Unconstrained entrepreneurs will pay the start-up cost and invest the optimal amount of capital, k*(w) + q £ A. Constrained entrepreneurs will pay start-up costs q £ A and invest some capital, although they would like to invest more. Some people have enough wealth to start a business but choose to work. These people do not have very much entrepreneurial skill relative to their wealth and this means that their start-up costs are greater than z(A,w). The function z(A,w) represents the marginal entrepreneur. It is found by fixing A and w and solving for the start-up cost that makes wage earnings equal to entrepreneurial earnings. This function is increasing and concave in wealth. Depending on parameters, however, this function may have a vertical portion and for sufficiently high levels of wealth it may be horizontal.

In this model, a financial innovation amounts to imagining that some or all of the agents borrow and lend with one another without restriction at an equilibrium interest rate. In this case there would be a critical value of wealth equal to the set-up cost, q , above which all decisions about whether to be an entrepreneur are independent of wealth. All entrepreneurs would be unconstrained. One could model the exogenous spread of financial institutions in this way.

This is a static model of occupational choice where households decide whether to work in the wage sector or to become an entrepreneur. The amount of capital that can be invested in entrepreneurial projects may be limited because credit markets are incomplete. In particular, households can borrow only a fixed multiple of their total wealth. The interest rate is fixed exogenously.

Households make their decision about whether to be an entrepreneur or to work in the wage sector by examining expected earnings in each sector and choosing the sector that delivers the highest expected earnings. For entrepreneurs, earnings are determined by their skill in business (q *) and by the amount of capital that they invest. Note that q * is negatively related to the measure for start-up costs in the LEB model, q . In the EJ model, when q * is higher, the entrepreneur is more talented. In the LEB model, when q is higher the entrepreneur is less talented and must pay a higher fixed cost in order to start a business. Entrepreneurial income is given by:

Y = q *ka e + r(A – k).

Here k is the amount of capital invested, e is a shock to the production function, a is a productivity parameter, and A is household wealth. Implicit in this representation of income is the possibility that income could be negative and that businesses could go bankrupt. More skilled entrepreneurs (higher q *) have higher total product and higher marginal product of capital at all levels of capital. Net savings (A – k) earns a riskless gross return of r. If A – k < 0 then the household must pay a gross interest rate of r on borrowed funds. For wage workers, total earnings are given by w + rA, where w is the wage rate.

Households can borrow up to a fixed multiple of their wealth to supplement their own investment in an entrepreneurial venture. The maximum amount that a household can borrow is given by:

(l - 1)A,

where in the model, l > 1. The borrowing limit in turn determines the maximum amount that the household can invest in the business as a function of their wealth. The maximum amount that can be invested in the business is given by A + (l - 1)A = l A. Therefore, investment, k, must satisfy:

0 £ k £ l A.

The amount of capital that an entrepreneurial household would like to invest is determined by entrepreneurial skill, the interest rate and by the productivity parameter a . Let k* represent the optimal amount of capital to invest in the business. In the absence of liquidity constraints, the optimal amount of capital to invest in the business is given by:

k* = (q *a /r)^{1/(1-a )}

Notice that in the absence of liquidity constraints, the optimal amount of capital to invest in the business is unaffected by household wealth. This is also true in the LEB model. However, unlike the LEB model, the optimal amount of capital to invest in the business will depend on the potential entrepreneur’s business skill, q *.

Higher skill entrepreneurs will want to invest more capital. This means that liquidity constraints are more likely to be binding for higher skill individuals since business skill and capital are complements. Holding household wealth constant, higher skill individuals are more likely to want to invest amounts of capital that exceed l A, the maximum dictated by the borrowing limit. Entrepreneurs will be unconstrained only if their business skill is low enough. That is if business skill satisfies:

q * £ (r/a )(l A)^{1 - a} .

The decision about whether to become an entrepreneur depends on the difference between entrepreneurial income and income in the wage sector. For unconstrained entrepreneurs (k* <l A), entrepreneurial income will be equal to:

Y^u = q *(k*)a e + r(A – k*).

Unconstrained households will choose to open a business if q *(k*)a e + rk* > w. Notice that in the absence of liquidity constraints the decision to become an entrepreneur is independent of household wealth.

The earnings of constrained entrepreneurs will be:

Y^c = q *(l A)a e + r(A – l A ).

Constrained households will open a business if q *(l A)a e - rl A > w. In this case the decision to become an entrepreneur does depend on household wealth. Holding other factors constant for a constrained household, an increase in household wealth increases entrepreneurial income and the likelihood that a household becomes an entrepreneur.

The predictions of this model are summarized in Figure 2, which describes who becomes a wage worker, who becomes a constrained entrepreneur and who becomes an unconstrained entrepreneur as a function of business skill, q *, and household wealth, A. Holding entrepreneurial skill constant, and moving along the wealth axis, we see that at low levels of wealth, individuals will choose to be wage workers. If wealth increases, at sufficiently high levels of entrepreneurial skill, the same individual will open a business, although the amount of capital invested in the business will be constrained by household wealth. If wealth increases even more, the individual will move to operating the business at an efficient level of capital. Further increases in wealth will not change entrepreneurial earnings.

In this model, an improvement in the financial system is equivalent to an increase the parameter l , the multiple of wealth that the household can borrow. As in Paulson (1997), one could imagine that l varies by institution if different institutions have different technologies for catching and/or punishing renegers. It might also be reasonable to allow l to vary by village or region to reflect differences in the number and type of financial institutions that are available in different areas. Allowing l to vary across individuals would be one way to incorporate the impact of individual characteristics on the supply of credit.

This is a dynamic model of occupational choice with limited information and incentives in a classic moral hazard problem. Each household has a technology for producing output q from its own effort z and from capital k. This technology is written as P(q | z, k), the probability of achieving output q given effort z and capital k. Following Aghion and Bolton (1996), k is equal to one and represents investment in a relatively large indivisible project. Output q can be either high, q = 2, if the project is successful or low, q = 0, if the project fails. The probability that output will be high is equal to the level of effort that is exerted, if one unit of capital is invested. If no capital is invested, output is zero. That is:

z = P(2 | z, k = 1) and 0 = P(2, z, k = 0)

In other words, the likelihood of success in increasing one for one with effort, if the firm is capitalized, otherwise the firm does not operate and there is no output. Effort z is assumed to lie in the interval [0,1]. We also consider generalizations of this technology below.

Like all of the models that we consider, the household is risk neutral, with a utility function that is linear in end of period wealth, t . However, utility is decreasing in effort, although effort is less costly in utility terms for more talented entrepreneurs. Household utility is given by:

U(t , z) = t - z²/q *,

where, as in the previous models, q * represents entrepreneurial talent. Talented, high q * households are able to produce more effort for a given level of disutility than their less talented counterparts. Like z, q * is assumed to lie in the unit interval [0,1].

Households begin the period with wealth, A, that has been inherited from the previous generation. This wealth can be invested in the household’s business or it can be reallocated to other households via lending. The household can also borrow if it does not have enough wealth to capitalize its business. The amount of borrowing is given by 1 – A. Borrowing and lending are subject to the market clearing interest rate, r.

In effect, all household funds are put on deposit at the beginning of the period, and then households who become entrepreneurs borrow what they need for investment. Repayment of borrowed funds may depend on realized output, q. At the end of the period, wealth (t ) is either consumed or saved. Some fraction, s, of final wealth is passed along to the next generation and the sequence is repeated. The dynamics of this model generate growth with increasing inequality, but that is not the focus here.

The focus of the model, and of our analysis, is the incentive contract for households who borrow. Since effort, z, is unobservable, households must be given an incentive to be diligent and exert effort equal to z. The incentive is created by letting household proceeds from running the business (t ) depend on output, q. One could let consumption be negative, so that loans are repaid even when output is 0, but this class of models realistically precludes that option. It is this lower bound on consumption that is the cause of the incentive problem, despite risk neutrality. Given an assignment of k and recommended effort, z, and any deviant effort, z¢ , the incentive constraint is given by:

The deviant action z¢ enters into the utility function on the right hand side. The ratio of probabilities on the right hand side rescales the probability of output q to take into account the deviation.

The linear quadratic nature of the model makes it possible to derive analytic solutions. Households whose initial wealth, A, is below the required level of investment, k = 1, who are able to borrow will receive a return of t = 0 if the project fails and receive a positive return if the project succeeds. In the case of success, the return is given by:

[1]

Under this contract, incentive compatible effort, z, is simply equal to (q */2)t .

The amount borrowed is equal to (1 – A) and both effort and the return to a successful project are increasing in this amount. The project return is increasing in the amount of funds borrowed in a strictly concave fashion. By the same logic, as wealth decreases effort also decreases and there is a minimum level of wealth A* below which projects will not be funded. A* is equal to:

A* = 1 - q */2r.

The intuition of this result is that the lender would have to take so much of the project return away from low wealth borrowers that these borrowers have little incentive to work hard. The only way to compensate for this is to increase the interest rate as wealth decreases, but at some level of wealth the market will clear with no borrowing. Above A* and below A = 1, Lehnert (1999) has computed examples in which households who face the interest rate implicit in the return function [1] would like to borrow more than their allocation (1 – A). These households could increase their utility if they could borrow more and commit to higher effort. However, since effort is not publicly observed, they are unable to credibly commit to this plan. In this sense, the model implies that firms who with wealth between A* and 1, who must borrow to finance investment, are constrained. Relaxing the constraint in this model would increase entrepreneurial effort, although capital investment would remain fixed at one unit. In contrast, in the Lloyd-Ellis and Bernhardt and the Evans and Jovanovic models relaxing constraints would lead to increased business investment. We use the term "constrained" to refer to firms who constrained for any reason.

As wealth increases, the household will reach the point where it does not need to borrow in order to invest k = 1 in their business. At that point, entrepreneurs face the full costs and benefit of their effort, the private information problem is not an issue and production is efficient. Regardless of entrepreneurial output, households with A > 1, earn r(A – 1) from their lending activities. Talented entrepreneurs are more likely to have successful firms, since they exert greater effort.

In Figure 3, we plot expected utility as a function of wealth, A. This reveals another contractual possibility, beneficial wealth lotteries. As emphasized in Lehnert’s paper, everyone whose wealth is below A* should pool their wealth and play a lottery where winning households get wealth of A^L. Winning households are able to enter into incentive compatible contracts with the lender and can become entrepreneurs. The probability of winning the lottery increases linearly with wealth as wealth approaches A*.

In summary, this model implies that households will choose their occupations depending on their wealth and their entrepreneurial talent. These choices are summarized in Figures 4A and B for the lottery and the no-lottery case, respectively. In the no-lottery case, households with initial wealth below A* will be inactive savers. In the case with lotteries, even households with wealth below A* will start businesses with some positive probability, depending on whether or not they win the lottery. Households with wealth between A* and 1 will start businesses and supplement their own investment with borrowed funds. These households are constrained in the sense that they would like to borrow more. Households with initial wealth that is greater than their investment needs of 1, will start businesses and save their surplus wealth. These households choose the optimal amount of effort and are unconstrained.

It is possible to generalize this model, especially if we resort to linear programs as Lehnert does. One generalization that would allow a more direct comparison of this model with the other two would be to introduce a variety of scales of investment. Risk aversion can also be introduced. A third innovation would make output conform to alternative production functions that map effort and capital into success probabilities. In some of these examples, investment will be a concave, increasing function of wealth, up to some critical level of wealth after which investment is constant. This investment would be partially financed through borrowing. Borrowing would decrease with wealth up to the critical level after which firms are fully self-financed and residual wealth is saved. In other examples, investment is decreasing monotonically with wealth. This happens when capital and effort are compliments. High wealth agents should not be asked to work hard and they are assigned less capital accordingly. At the low end of the wealth distribution, investment will still increase with wealth as in the other example. In that region, increases in wealth mean that households will face incentive contracts that require greater effort. The rewards and penalties that entrepreneurs receive when their projects succeed and fail in these examples are very revealing. Even with risk aversion, some variation in returns and project outcomes is information constrained efficient.

Because the Lloyd-Ellis and Bernhardt, the Evans and Jovanovic and the Aghion and Bolton/Lehnert models share many characteristics, we consider their implications together. First these three models help us address the important issue of whether or not liquidity constraints are an important feature of the rural and semi-rural Thai economy. If liquidity constraints are not an important factor in determining which households run businesses in the data, then according to these models, predictions of which households will run businesses will be unaffected by the levels of household wealth.

The second set of implications that we draw from these models comes directly from Figures 1, 2 and 4A. These figures make predictions about how the fraction of business owners and the fraction of these business owners who are constrained and unconstrained will vary with wealth and entrepreneurial talent. Holding entrepreneurial talent fixed, all three models imply that as wealth increases the percentage of business owners will increase. The rate of increase may be constant, decreasing or even degenerate to zero at some levels of wealth depending on the region and the graph. In all of the models, at very high levels of wealth, the rate of increase in the fraction of business owners will fall to zero. Overall, the models imply that the percentage of business owners will increase with wealth at a decreasing rate.

In addition, we should expect that the percentage of business owners who are constrained will fall with wealth, again holding talent fixed. This percentage should also fall at a decreasing rate as wealth increases. In both the LEB and the EJ models, the percentage of business owners who are constrained will decrease at a constant rate over some range. At sufficiently high levels of wealth, no one will be constrained and there will be no further change in the fraction of constrained business owners.

Another set of predictions comes from fixing wealth and changing entrepreneurial skill. If we fix wealth and increase entrepreneurial skill or equivalently decrease start-up costs, all three models predict that we will find more and more business owners.

So far the LEB, the EJ and the ABL models have the same implications. However, the models imply different things will happen to the percentage of constrained and unconstrained firms when we fix wealth and increase business talent. Looking at Figure 1, we see that in the Lloyd-Ellis Bernhardt model when we fix wealth and increase entrepreneurial talent, we may move from the region of constrained entrepreneurs to the region of unconstrained entrepreneurs. When we perform the same exercise for Figure 2, the EJ model, we do the opposite. As entrepreneurial talent increases, we move from the unconstrained region to the region of constrained entrepreneurs. The Lloyd-Ellis Bernhardt model implies that as the fixed cost falls for a given level of wealth, the percentage of constrained entrepreneurs will either be constant or decrease. In contrast the EJ model predicts that the percentage of constrained entrepreneurs will increase as talent increases when wealth is held fixed. In the ABL model, given that one has been able to start a business, changes in business skill have no impact on whether the household is constrained or not.

In the EJ model, entrepreneurial talent and capital are complements. The more talented you are, the more capital you would like to invest. This means that for a given level of wealth, more talented people are more likely to hit the borrowing constraint. In the LEB model, entrepreneurial talent is only important in determining the fixed cost of starting a business. When talent increases, the fixed cost goes down and constrained entrepreneurs are able to invest the freed up capital. For some entrepreneurs, the freed up capital will be enough to let investment grow to the unconstrained level. In the ABL model, as we have interpreted it, all firms who borrow to finance investment are constrained, regardless of business skill.

If one were willing to make assumptions about how entrepreneurial talent is distributed across individuals and about the characteristics of the joint distribution of talent and wealth, then all of the implications that we have discussed so far could be made more specific. Suppose, for example, start-up costs are uniformly distributed from 0 to 1 across individuals at every level of wealth. If this were the case, then the length of a vertical line segment on Figure 1 at a given level of wealth from the x-axis to the function z(A,w) should be exactly equal to the percentage of people who run businesses. Depending on where we fix wealth, the length of this line may indicate the percentage of constrained or unconstrained entrepreneurs, or some combination of them. Similar exercises can be done with Figures 2 and 4.

We can also use these models to develop predictions about how business investment will vary with wealth and entrepreneurial talent. By construction in the Lloyd-Ellis and Bernhardt model, if less talented individuals start business then they will have to make larger initial investments. The model also assumes that for a given level of entrepreneurial talent, start-up investment will be unchanged when wealth changes. However, the model predicts that average investment will go up with wealth as individuals with higher and higher start-up costs are able to leave wage work and start businesses.

For a given start-up investment, the LEB model predicts that marginal investment in the firm will increase one for one with wealth until the unconstrained optimal level of capital is reached. Further increases in wealth beyond this point will have no impact on investment. The Evans and Jovanovic model makes a qualitatively similar prediction. However, wealth increases in this model can have an even larger impact on investment for constrained firms because the increase in wealth is leveraged through borrowing. For a constrained firm at a given level of entrepreneurial skill, a one unit increase in wealth will lead to an increase in investment of l units (l is greater than one). As in the LEB model, the EJ model implies that increases in wealth will have no impact on investment once the unconstrained region is reached. In the Lehnert version of the Aghion and Bolton model, investment in the business is constant across the wealth distribution, by assumption.

The models also have implications for how savings and debt will evolve with wealth and entrepreneurial talent. Since the credit market is non-existent by construction in the Lloyd-Ellis and Bernhardt model, we concentrate on developing the implications of the Evans and Jovanovic and the Aghion and Bolton/Lehnert model for net savings. Both of these models imply that savings will equal wealth so long as households choose to work in the wage sector rather than open a business. If wealth increases, the same household may now find it profitable to open a business, although in the EJ model the scale of the business may be constrained by wealth. At the point of transition from wage work to running a constrained business, net savings will jump from A to –(l - 1)A in the EJ model, as the household borrows the maximum amount and invests it in the new enterprise. In the ABL model, savings will go from A* to -(1 – A*) at the point of transition from wage work to entrepreneurship. Both models imply that households who do not own businesses should be net savers while constrained business owners should be net borrowers.

As long as the household is operating the business in the constrained region, further increases in wealth will increase borrowing in the EJ model and decrease borrowing in the ABL model. A one unit increase in wealth allows borrowing to increase by l - 1 units for constrained businesses in the EJ model. This is interesting because it suggests that household businesses with greater amounts of debt will be more likely to be constrained. In contrast, the ABL model implies that increases in wealth will reduce borrowing by an equivalent amount.

In the EJ model unconstrained businesses may finance some investment in the firm by borrowing. In the ABL model, firms are unconstrained only when they can self-finance all firm investment. For unconstrained businesses in the EJ model, increases in wealth have no impact on total investment in the firm, but they may affect how leveraged investment in the firm is. Consider an increase in wealth for the firm that is just on the border between being constrained and unconstrained, where k* = l A. When this household receives an additional unit of wealth they can reduce the amount of borrowed funds that are invested in the firm by one unit, or equivalently increase savings by this amount. In any case, in all of the models, net savings will increase one for one with wealth for unconstrained firms the same way it does for wage workers. In the ABL and LEB models net savings will always be positive for unconstrained firms.

In summary, all three models imply that the percentage of business owners will increase with wealth and with talent. All three models also imply that the percentage of constrained business owners will decrease with wealth. Holding wealth constant, the EJ model implies that the percentage of constrained entrepreneurs will increase with talent and the LEB model makes the opposite prediction. The ABL model suggests that talent will have no impact on the fraction of entrepreneurs who are constrained. Both the LEB and the EJ models predict that entrepreneurial investment will rise with wealth for constrained firms. All of the models imply that savings will increase one for one with wealth for non-business owners and for unconstrained business owners. For constrained firms, the ABL model implies that borrowing will decrease with wealth, while the EJ model suggests that the opposite will happen.

6. Empirical Results and Discussion

In this section we develop and discuss empirical counterparts to the implications of the models that were discussed above. The first implications that we evaluate are those that are summarized in Figures 1, 2 and 4A: how the percentage of wage workers and constrained and unconstrained business owners varies with wealth and entrepreneurial talent. In order to do this, we need to come up with some observable proxy for entrepreneurial talent. The proxy we use is education. While education is certainly not a perfect indicator of entrepreneurial talent, it is likely to be positively correlated with business skill.

We also need to come up with appropriate measures of wealth. The wealth variable in the models is beginning of period wealth, that is wealth prior to starting a choosing an occupation. As an empirical counterpart to this variable, we use wealth six years prior to the survey. The items that are included in this measure are: household and agricultural assets and titled and untitled land. We do not include the value of business assets that the household may have owned six years ago. By using past, rather than current wealth, and by excluding business assets we hope to avoid issues of endogeneity. The endogeneity problem that we are concerned with is the following: wealthier people are more likely to start businesses and business owners have higher earnings than wage workers, which allows business owners to become still richer. In this scenario, current wealth captures both the cause and the effect of having been able to start a business in the past. Since 60% of businesses were founded in the past six years, our measure of past wealth captures conditions prior to opening the business for most households.

In order replicate Figures 1, 2 and 4A using the data, we also need to be able to tell whether or not businesses are constrained. We consider households who answered yes to the question "Would your business be more profitable if it were expanded?" to be constrained. There are several issues surrounding the use of this measure. First, the question refers to current not past conditions. Second, it is unclear to what extent the respondents’ interpretations of the question match the model definition of constrained. Third, the question is more consistent with the notion of constraints that is important in the Lloyd-Ellis and Bernhardt and the Evans and Jovanovic models than the concept of constrained that emerges from the Lehnert model. The responses to follow-up questions about why the household did not expand the business suggest that there is at least a rough correspondence between the theoretical and the practical definitions of being constrained. More than 55% of the constrained households said that the reason that they did not expand was because they lacked funding. Another 20% said that they lacked land or labor (see Table 1D).

Table 2A summarizes the percentage of non-business owners, the percentage of constrained business owners and the percentage of unconstrained business owners by the education of the head of household and wealth. We split years of schooling into three groups: 0 – 3 years, 4 years and 5 – 16 years. About 60% of the heads of household in the survey have 4 years of schooling, which was the statutory minimum at the time the cohort that most of them belong to cohort went to school. Wealth is divided into quartiles. The division depends on the sample under consideration. So, for example, only northeastern households were taken into account in determining the quartiles for the Northeast. The figures in this table summarize the key features of the data along the lines suggested by the models without making a commitment to a particular statistical model.

All of the models predict that, holding business skill fixed, as wealth increases the percentage of business owners will increase. This prediction appears to be born out in the data. Comparing the percentage of business owners in the lowest wealth category to the percentage of business owners in the highest wealth category, we find that the percentage of business owners almost doubles for each region and for all education groups. The percentage of business owners increase from 11% to 22% in the Central region and from 3% to 6% in the Northeast for the lowest education group. For people with four years of schooling, the percentage of business owners goes from 22% to 43% in the Central region and from 8% to 17% in the Northeast. For the group with 5 – 16 years of schooling, the percentage goes from 24% to 50% in the Central region and increases from 21% to 34% in the Northeast.

In addition to predicting that business ownership will increase with wealth, the models also imply that, holding entrepreneurial skill constant, the percentage of constrained business owners should fall as wealth increases. We can examine this implication by comparing the percentage of business owners who are constrained in the lowest wealth category to the percentage who are constrained in the highest wealth category. Here the results are more mixed. In the Central region, the percentage of business owners who report that they are constrained increases from 27% to 59% from the lowest to the highest wealth quartile for heads with 0 – 3 years of schooling. For heads with have four years of schooling, the percentage of constrained business owners decreases slightly: from 50% to 47%. For the most educated group, the percentage of constrained business owners increases from 58% to 66%. In the Northeast, the results seem to line up a bit better with the models’ predictions. For heads with 0 – 3 years of schooling, the percentage of constrained business owners decreases from 100% to 50% from the lowest to the highest wealth quartile. For those with four years of schooling the percentage goes from 88% to 71%. For the most educated group, the percentage of constrained business owners increases from 38% to 59%.

All of the models imply that, at least for some wealth levels, the percentage of businesses will increase when business skill increases. We can get some indication if this is the case by comparing the percentage of households who have business in the lowest education group to the percentage who have businesses in the most educated group, for each wealth quartile. Here the results are very much in line with the predictions of the models. In each region and for every wealth quartile, the percentage of households who run businesses is much higher in the most educated group compared to the least educated group. For example, in the Central region, 19% of households in the second wealth quartile whose heads have 0 - 3 years of schooling run businesses, compared with 38% of the households with 5 – 16 years of schooling. The same comparison for the Northeast is 6% v. 25%.

So far we have looked at implications that are shared by all of the models. However, when wealth is fixed and business skill increases, the models make different predictions. The ABL model predicts that there will be no change in the number of constrained and unconstrained business owners. The LEB model implies that the percentage of constrained entrepreneurs will be either constant or decrease when skill increases and wealth is held constant. In contrast, the EJ model predicts that the percentage of constrained business owners will increase with business skill. We can evaluate these predictions by comparing the percentage of constrained business owners in the lowest education category to the percentage of constrained business owners in the highest education category, for a given wealth quartile. In the Central region, the percentage of constrained business owners increases for two of the four wealth quartiles. For the lowest quartile the percentage jumps from 27% to 58% and for the highest quartile it climbs from 59% to 66%. The percentage of constrained business owners falls in the other two quartiles: from 63% to 38% in the 2^nd wealth quartile and from 48% to 46% for the 3^rd quartile. In the Northeast, the percentage of constrained business owners increases in three of the four wealth quartiles. The percentage increases from 57% to 60% in the 2^nd quartile, from 57% to 65% in the 3^rd and from 50% to 59% in the highest wealth quartile. In the lowest wealth quartile, the percentage of constrained business owners falls from 100% to 38% when we move from the least educated to the most educated group. So for five of the eight groups, the percentage of constrained business owners increases with education, although for some groups the increase is fairly modest. Overall these findings seem to offer some evidence in favor of the EJ model’s assumption that business skill and capital are complements.

The evidence presented in Table 2A and discussed above suggests that all three of the models capture the key features of the data: business ownership increases with wealth and with education. This is exemplary given how stylized the models we consider are. When the model predictions differ, as they do in their predictions about how the percentage of constrained entrepreneurs will change with business talent, the evidence seems to favor the Evans and Jovanovic assumption that talent and capital are complementary. Some aspects of the data cannot be reconciled by any of the models. For example, we sometimes find that the percentage of constrained entrepreneurs increases with wealth, and we also do not find evidence that the rate of increase in business ownership decreases with wealth. Both of these findings could be validated if we assumed that there are non-convexities in production.

So far we have examined the implications of the models without making any statistical assumptions. In the next set of tables, Tables 3A, 3B, 4A and 4B we estimate probit models of who becomes and entrepreneur and who becomes a constrained entrepreneur. In the first set of estimates (Tables 3A and 3B), the dependent variable is equal to one if the household currently runs a business that was founded in the last 5 years. The explanatory variables include characteristics of the household head that may be indicators of business talent: age, age squared, and years of schooling. There also variables that control for the amount of household labor that is available: the number of adult males, the number of adult females and the number of children living in the household. Issues of labor supply are largely outside the discussion of the models under consideration here. One way to interpret their inclusion in the estimates is simply as a control, so that the variables of interest can be related to the implications of the models.

We also include measures of household wealth, being careful to take issues of endogeneity into consideration. The wealth variables are: the value six years ago of household, agricultural and land assets that the household owned then. This figure excludes the value of land that was inherited between six and ten years ago. The value of inheritances received between six and ten years ago is a separate indicator of household wealth prior to starting the business. Inheritances provide a useful source of truly exogenous variation in household wealth. It is possible that inheritances are anticipated. If that is the case and we find that they are positively associated with starting a business, then this is especially strong evidence of liquidity constraints. In the absence of liquidity constraints, we would expect households to be able to borrow against anticipated inheritances. We also include the interaction between past wealth and inheritances. If the likelihood of starting a business increases with wealth at a decreasing rate, as the models predict, the coefficient on this variable should be negative.

The set of explanatory variables that we use in Table 3A is meant to correspond loosely to the Lloyd-Evans and Bernhardt model. In their model, there is no credit market. In the data, however, households do borrow and lend. We control for credit market availability by including measures of whether or not the household was a member or a customer of various financial institutions in the past. Like the labor supply variables, we include these variables so that we can appropriately interpret the talent and wealth variables in the spirit of the Lloyd-Evans and Bernhardt model. The credit market variables provide an indication of the average probability that patrons of the various institutions will start businesses.

Our findings for the whole sample are largely consistent with the implications of this model (and the others). An additional year of schooling increases the likelihood that a household will start a business by about 0.6%. Controlling for past wealth, receiving an inheritance of 1,000,000 baht increases the likelihood that the household will start a business by 6%. This is an increase of 46% above the observed percentage of households who have started a business in the past five years. Past wealth is also an important indicator of the probability of starting a business. The likelihood of starting a business would increase by 0.8% if past household wealth went up by 1,000,000 baht. While the sign on the coefficient for interaction between wealth and inheritance is negative, it is not significant. This finding is consistent with the lack of evidence in Table 2A that entrepreneurship increases at a decreasing rate with wealth.

The estimates for the Northeast and the Central region suggest that entrepreneurial talent and wealth have a different role in determining which households will open businesses in the two regions. In the Northeast, education is important and wealth is not. In the Central region the opposite is true. An additional year of schooling in the Northeast increases the likelihood of opening a business in the next five years by 1%. This would represent a 12% increase in the percentage of business owners in the Northeast. In the Central region, the education coefficient is not significantly different from zero. In the Northeast, the wealth variables are all insignificant. In contrast, a 1,000,000 baht increase in inheritance in the Central region would raise the likelihood of opening a business by 7%, a 41% increase. One interpretation of these findings is that entrepreneurial talent is the scarce resource in the Northeast. This is consistent with Thai migration patterns. There has been substantially more out-migration from the Northeast than from the Central region. Migrants tend to be younger and more educated (and possibly more talented in business) than non-migrants.

In Table 3B we augment the set of explanatory variables so that they correspond roughly to the models with credit, Evans and Jovanovic and Aghion and Bolton/Lehnart. These models imply that in addition to providing additional investment funds, increases in wealth also relax borrowing constraints. Although an increase in wealth relaxes constraints in both models, in the EJ model an increase in wealth leads to more borrowing while in the ABL model increases in wealth allow the household to borrow less. In any case, we add interactions of wealth with the measures of past use of financial institutions in order to capture the effect of wealth on credit constraints. If the models implications are borne out, then these variables will increase the likelihood of opening a business.

The findings suggest that the effect depends on the type of institution. If the household was a customer of the BAAC, for example, a 1,000,000 baht increase in wealth raises the likelihood of starting a business by 1% through the effect of that institution. In contrast, if the household was a customer of a moneylender, the same increase in wealth would decrease the likelihood of opening a business by 2% through the moneylender effect. This suggests that there may be an important role for selection in credit markets. There is no impact on the likelihood of starting a business from the interaction of wealth with the variables that indicate whether the household was a customer of a commercial bank, a village institution or a ROSCA. Looking at the regional results, we find that none of the interaction terms are important in the Northeast. The results in the Central region are similar to the overall results. It is also important to note that, compared to Table 3A, adding these variables did not change the magnitude or the significance of the coefficients on the other variables. Wealth and inheritances are still an important predictor of who will open a business in the Central region. In the Northeast, the wealth variables are insignificant, but additional years of schooling significantly increase the likelihood of opening a business.

Tables 4A and 4B use the same sets of variables to predict who will be a constrained business owner. These tables provide some evidence that being constrained is related to wealth. In the specifications which use the whole sample, receiving a 1,000,000 baht inheritance six to ten years before the survey would decrease the likelihood of being constrained at the time of the survey by about 7%. The magnitude is roughly the same in the Central region. In the Northeast, however the same inheritance would decrease the likelihood of being constrained by about 275% to 292% (depending on the specification). The coefficient on inheritance is significant at the 10% level. In the specification in Table 4A, none of the other wealth variables are significant. Education is also insignificant in the all of the estimates. To the extent that years of schooling captures entrepreneurial skill, this finding contradicts the EJ model which implies that more skilled individuals are more likely to be constrained.

Increases in wealth may also reduce the likelihood of being constrained by freeing up borrowing constraints. This effect is included in the estimates in Table 4B via the variables that interact wealth with membership in various financial institutions. There is some evidence that the joint effect of wealth and membership in a village organization can increase the likelihood of being constrained. For households in the Central region, holding other things fixed, a 1,000,000 baht increase in wealth will increase the likelihood of being constrained by 3% through this effect. Again the evidence in the data suggests that selection in the credit market may be important. In contrast, if you are the customer of a formal financial institution, the same increase in wealth will decrease the likelihood that you are constrained by 2% through the interaction effect. There is some weak evidence that the interaction of wealth with being the customer of a moneylender reduces the probability of being constrained. Note, however, that the direct effect of being a customer of a moneylender is to increase the probability of being constrained.

These estimates of the probability of opening a business and the probability of being constrained suggest liquidity constraints are important, especially in the Central region. In addition, education is an important determinant of who will open a business in the Northeast. Like the models imply, the probability of being constrained is decreasing in wealth. The results also suggest that selection is important in two respects. Selective out-migration may limit the range of business skill in an area, like it appears to have done in the Northeast. In addition, selection in financial markets would help to explain the different effects that we find in the way wealth interacts with different financial institutions. Screening of clients may play an important role in the rural credit markets we are considering. So far, in the models we have considered, the credit market screens only on wealth but not on other characteristics.

In addition to their predictions about who will be a constrained or an unconstrained business owner, the models that we have been considering also make predictions about how current net savings and initial business investment are affected by wealth and business skill. Tables 2B and 2C summarize median net savings and initial business investment by wealth quartiles and by years of schooling. Net savings is equal to financial savings plus the value of savings in goods plus the value of loans owed to the household minus household debt. The models predict that net savings will be positive and increasing with wealth for non-business households. For business households, the models predict that net savings may be negative. These predictions mirror the overall pattern we see in the data. For non-business owners, net savings is –500 for the lowest wealth quartile, at higher levels of wealth it is equal to zero. For business owners net savings is consistently negative across all wealth quartiles.

The predicted relationship between net savings and wealth depends on the model under consideration and whether or not the business is constrained. In the ABL model, constrained business owners will borrow and unconstrained business owners will not. In the Central region, we see precisely this relationship. Net savings for constrained business owners is negative regardless of where in the wealth distribution they are. For unconstrained business owners, net savings is zero for the lowest wealth quartile and ranges from 98 to 1400 baht in the other quartiles. In the Northeast, both constrained and unconstrained business owners have negative net savings, and sometimes the unconstrained group borrows more than the constrained group and sometimes it is the other way around.

The EJ model and the ABL models have different predictions about how net savings will change with wealth for constrained business owners. In the EJ model, borrowing will increase with wealth for constrained business owners. In the ABL model, the opposite is true, borrowing will decrease with wealth for constrained business owners. We can evaluate this prediction by comparing the net savings of the poorest group of constrained business owners to the net savings of the richest group. In the Central region, there is little change in net savings across the wealth quartiles for constrained business owners. For the lowest group, median net savings is –20,218 baht and for the highest group median net savings is –19,843 baht. In the Northeast, the net savings (borrowing) of constrained business owners seems to decrease (increase) as predicted by the EJ model. For the poorest group, median net savings is –361 baht. Median net savings is negative 7,010 baht for the richest group.

All of the models imply that net savings should increase with wealth, or equivalently that borrowing should decrease with wealth, for unconstrained business owners. By comparing the net savings of the poorest group of unconstrained business owners to the net savings of unconstrained business owners in the highest wealth quartile, we can examine this prediction. Looking at the Central region, we find evidence in favor of the model: median net savings goes from 0 to 1,400 baht across the wealth quartiles. In the Northeast, we find the opposite pattern. Net savings for unconstrained business owners falls from 4,500 baht for the lowest group to –5,200 baht for the highest group. This pattern is inconsistent with all of the models. This pattern would make sense in the context of the EJ model if these businesses were in fact constrained, even though they have reported that they are not.

The Lloyd-Ellis and Bernhardt model implies that initial investment in the business will be decreasing with entrepreneurial talent. Less talented individuals will need to invest more than their more talented counterparts who start businesses. In contrast, the EJ model implies that highly skilled workers would like to invest more, although they are more likely to be constrained. On average, however, this model implies that investment will increase with talent. We can examine these implications by looking at how initial business investment varies with wealth and education. This information is summarized in Table 2C. The pattern of initial investment and education does not favor the LEB model. Instead of finding that investment decreases with education, we find the opposite. In the Central region, median investment increases by 15% from the lowest education category to the highest. In the Northeast, the increase is even more dramatic; investment goes up by 56%. This finding offers more evidence that capital and entrepreneurial talent are complements, as the EJ model assumes.

The LEB model also implies that average investment will go up with wealth, as individuals with higher and higher start up costs are able to start businesses. The EJ model makes a similar prediction, although the mechanism is different. Median business investment for firms in the lowest wealth quartile is 25,985 baht in the Central region, compared to 51,401 baht for firms in the highest wealth quartile. In the Northeast, median investment increases from 14,118 baht to 34,953 baht. Again, this pattern is consistent with the broad predictions of the models.

Both the LEB and the EJ models also imply that total investment in the firm will increase with wealth until the unconstrained optimal level of capital is reached. This means that investment will increase at a decreasing rate. We can examine the rate of increase by comparing the percentage change in investment when we move from the 1^st to the 2^nd wealth quartile to the percentage change when we move from the 3^rd to the 4^th quartile. In the Central region, median investment grows by 15% from the 1^st to the 2^nd quartile. From the 3^rd to the 4^th quartile, the growth in investment is 36%. In the Northeast, we see a similarly puzzling pattern. From the 1^st to the 2^nd quartile, median investment declines by 25% and from the 3^rd to the 4^th quartile median investment grows by 84%. The evidence on the rate of increase in business investment with wealth does not support either of the models. As we have seen from other perspectives, these findings would be consistent with regions of increasing returns to scale in production.

In addition to comparing median investment across wealth and education categories, we can also evaluate investment’s relationship with the important variables more systematically. Table 5A presents regression estimates of the log of initial business investment as a function of some of the same variables we have discussed earlier: age and education, the number of adult males, females and children in the household, wealth and inheritance. If liquidity constraints are important in the sense of the EJ and the LEB models, then wealth will have a significant positive effect on investment. The inheritance variable is insignificant in all of the regressions. However, past wealth is an important predictor of investment in the Northeast. If past wealth were to increase by 10%, business investment would go up by almost the same amount, by 9.3%. In the Central region, all of the wealth variables are insignificant. Education is significant and positively related to business investment in both regions. In both the Central region and the Northeast, an additional year of school increases business investment by about 13%.

These estimates of investments provide an interesting counterpoint to the probit estimates of who will start a business. In the Northeast, education, but not wealth, is an important indicator of who will start a business. In contrast, both education and wealth predict the scale of business in the Northeast. These results suggest that business start-ups in the Northeast are not liquidity constrained on the extensive margin, since wealth does not affect the likelihood of starting a business. On the other hand, business skill as proxied by education does seem to be in short supply. Given that households are running a business, the scale of the enterprise is limited by wealth in the Northeast. In the Central region we find almost exactly the opposite results. In the Central region, the decision to start a business is a limited by household wealth. However, wealth does not seem to affect the amount that entrepreneurial households invest in their businesses.

Given the relative agreement between the implications of the Evans and Jovanovic model and the patterns in the data, it is worthwhile to try to rationalize these findings in light of this model. These findings suggest that households in the Northeast and the Central regions also occupy different regions of Figure 1. People in the Central region seem more likely to have "average" business skill, q *. At low wealth levels they are constrained wage workers, but if wealth increases, they are able to open business and move immediately to the unconstrained business owner category. This is consistent with the findings that wealth is an important determinant of whether households have a business in the Central region, but not how much they invest in the business. Most people in the Northeast have less business skill and would work in the wage sector regardless of their wealth. However, there are a few people in the Northeast who have enough entrepreneurial talent to leave wage work and open a business. These businesses are operated at a scale that is constrained by wealth. This would be consistent with the findings that wealth is not an important determinant of who opens a business in the Northeast, but that wealth is important in determining the amount of investment in the business. In order to fit into the model, these people would have to have more business skill, on average, than their counterparts with businesses in the Central region.