TY - JOUR
AU1 - Artunç,, Cihan
AU2 - Guinnane, Timothy, W
AB - Abstract Recent research disputes the view that the joint-stock corporation played a crucial role in historical economic development, but retains the idea that the costless firm dissolution implicit in non-corporate forms deterred investment. A multi-armed bandit model demonstrates the benefits of costless dissolution in an environment where potential business partners are not fully informed. Experimentation creates a spike in dissolution rates early in firms’ lives, as less productive matches break down and agents look for better matches. Many of the better matches adopt the corporate form, whose higher dissolution cost functions as a commitment device. We test the model’s predictions using firm-level data on 12,000 enterprises established in Egypt between 1910 and 1949. The partnership reflected a trade-off between committing to a partner and sorting into potentially better matches, fostering the formation of more productive enterprises (JEL D21, D22, L26, N15, O16). What is the relationship between the legal organization of business firms and economic development? The rules for business enterprises affect firm longevity and capital accumulation. Partnerships do not exist independent of their members. The enterprise can be dissolved at any moment if a partner dies or withdraws. The corporation, on the other hand, exists separately from its owners and has locked-in capital (Blair and Stout 2005; Stout 2005; Dari-Mattiacci et al. 2017). An influential literature views the corporate form as critical in Europe’s and the United States’ successful industrialization (e.g., Chandler 1977; Cochran 1977). These works argue that because they are at-will, partnerships could not undertake large-scale, long-term investment. The corporate form’s role is held to be especially significant for emerging economies and late industrializers. Countries that had high barriers to incorporation, the argument goes, made it difficult to mobilize capital beyond the relatively modest personal wealth of individuals (Owen 1991; Kuran 2011). Recent research disputes the idea of a universally superior corporate form, but still views costless dissolution as a disadvantage to partnerships (Lamoreaux and Rosenthal 2005; Guinnane et al. 2007). Partnerships, however, were long the dominant form of multi-owner enterprises, even after the corporation became easy to create. We propose a novel interpretation of the partnership’s popularity: its costless dissolution encourages beneficial experimentation. We formalize the problem with a theory of company formation and dissolution, then provide empirical evidence consistent with the model’s predictions. This empirical exercise rests on a novel firm- and owner-level dataset composed of some 12,000 partnerships and corporations established in Egypt between 1910 and 1949. Our data allow us to make two distinct empirical contributions. First, because we have information on owners and their history in earlier enterprises, we can focus on the implications of owner matches for firm success. Second, the existing literature on firm dynamics pertains to recent experience in developed countries. The Egyptian data make a rare contribution to the literature on firms in developing countries. Most data on firms have limitations; they might be restricted to manufacturing establishments (e.g., Lee and Mukoyama 2015), exclude firms with self-employed owners (the recent literature that relies on the US Longitudinal Business Database, e.g., Decker et al. 2014), or miss certain legal forms such as single proprietorships (Guinnane and Martínez-Rodríguez 2018). Our Egyptian dataset lacks sole proprietorships, but does include firms from across the economy, including those run by the self-employed. Extensive robustness checks demonstrate that other data issues do not drive our main results. In our theoretical model, costless dissolution allows entrepreneurs to experiment with different partners until they sort into matches with the “right” partner. Only 75% of Egyptian partnerships in our data lasted at least 2 years, compared to 96% of corporations. For many ventures, high dissolution cost from the start would have locked partners into unproductive ventures or deterred formation of ventures in the first place. For other partners whose joint productivity is sufficiently high, however, the partnership’s ease of dissolution discourages larger investment and cooperation because agents can deviate from investment plans, dissolve the firm, and re-match with a new partner. The corporation’s lock-in feature offers these firms an option to commit. This article thus provides a new way to think about enterprise form adoption and firm longevity by taking seriously the trade-off between experimentation and commitment. We start with a model based on a multi-armed bandit framework. Agents match to one another to produce some surplus in each period. Match success rate depends on the match quality, which the partners do not observe but on which they share a common prior. The enterprise’s observed success rate informs the partners of the actual match quality. If partners share pessimistic posteriors, they dissolve the firm and move on to new partners. If partners believe the match quality is sufficiently high, they remain in their current match and provide increased investment or effort to the firm. At this point, however, the ease of dissolution becomes a hindrance because it encourages each partner to free ride on the other’s effort and then re-match. Thus, partnerships that survive the initial trial period find it profitable to “tie the knot” and incorporate in order to induce cooperation. Empirical analysis of the Egyptian multi-owner enterprises supports the model’s central prediction. Because of the learning mechanism, the risk of a firm’s dissolution differed significantly based on the underlying ownership structure. Partnerships suffered particularly high rates of attrition within the first few years of their establishment. Conditional on surviving this period, partnership failure rates dropped and became more uniform. The corporations’ risk of dissolution, on the other hand, remained fairly constant throughout the first few years of existence. Our empirical results support the view that dissolution reflects experimentation. Firms that contain partners who have been in business together before do not dissolve as quickly, because these firms have prior knowledge about match quality. Firms with family members also endure. Prior business experience with a partner in another firm provides information on match quality. But this effect does not hold for family members. Two brothers, for example, have considerable information on each other prior to starting a firm together. Partnerships dissolve most rapidly when their owners appear to have little prior knowledge of each other. This article contributes to a rich literature on the impact of institutions on economic performance. Acemoglu and Johnson (2001) and Acemoglu et al. (2002) are seminal papers on the institutional origins of contemporary economic development (see Acemoglu et al. 2005 for a review). A more recent literature uses within-country variation to explore the long-run effect of specific institutions (for instance, Banerjee and Iyer 2005; Dell 2010; Michalopoulos and Papaioannou 2013). We take an even more micro-approach by focusing on a specific legal institution—the menu of enterprise forms—to understand and document the learning mechanism underlying firm formation and dissolution dynamics. Chandler (1977) convinced many that the corporation held the key to economic development. The corporation indeed confers important advantages to large-scale enterprises that require investment with long maturity. Such advantages include legal personhood, capital pooling, limited liability, entity shielding, and concentrated management (Hansmann and Kraakman 2000). Economic historians, however, have challenged Chandler’s view by showing that partnership forms remained popular in many European economies (Guinnane et al. 2007), and even in the United States (Hilt and O’Banion 2009), despite the joint-stock corporation’s wide availability. The partnership’s popularity waned only after the introduction of the private limited liability company, which combined some of the corporate form’s benefits, like legal personhood, with partnerships’ contractual flexibility.1 This debate, however, has overlooked two points. First, Chandler’s recent critics have also stressed that the partnerships’ low dissolution costs had detrimental implications for enterprise growth. This work has not explored the point emphasized here that the partnership offered entrepreneurs an advantage: a way to experiment with potential partners at relatively low cost. Second, nearly all of the literature on historical enterprise forms has focused on the experience of Western Europe and the United States. But the corporate form could play a distinctly important role in the development of late-industrializers. Nicholas (2015) shows that joint-stock corporations enjoyed wide popularity in Japan and accounted for most of the country’s aggregate profits before World War II. Gregg (2015) shows that the corporation helped early 20th-century Russian manufacturing establishments raise capital more easily and achieve larger scale. The corporate form’s potential role in fostering development has recently received attention in the context of the Middle East. Kuran (2011) argues that the restrictive menu of enterprise forms in the Middle East underlies the region’s economic stagnation. He maintains that egalitarian and inflexible inheritance rules in Islamic law exacerbated the problem of untimely dissolution by requiring joint ownership of a partner’s assets for his inheritors. Incorporation would have allowed investors to avoid these disadvantages. Thus, in his view, high barriers to incorporation effectively prevented long-term, large-scale investment. Our model and empirical findings support this concern. As the model below highlights, some valuable firms cannot sustain high productivity without commitment, and commitment requires costly dissolution. Our empirical work also demonstrates the fragility of the partnership in the region Kuran studies. We differ, however, in stressing that the rapid dissolution of partnerships also offered benefits. The related economic history literature largely focuses on the choice of enterprise form. Guinnane and Martínez-Rodríguez (2018), for example, ask what types of firms preferred the corporation to partnerships in Spain, and consider the role of a new private limited liability form introduced in 1920. Other papers such as Guinnane et al. (2007), Abramitzky et al. (2010), Hilt and O’Banion (2009), or Gómez-Galvarriato and Musacchio (2006) study similar questions. This article extends the focus on legal forms to consider the longevity of enterprises organized in different ways.2 This article also contributes to a large literature on firm dynamics as well as a more specific theoretical and empirical literature on partnerships and matches.3 The literature on firm entry, exit, and lifecycle focuses on post-war developed economies, stressing establishments in the manufacturing sector (e.g., Lee and Mukoyama 2015). Bartelsman et al. (2004) identify several important empirical regularities. One regularity pertains to our empirics: the probability of an establishment’s (or firm’s) survival increases with its age (Dunne et al. 1988,; 1989; Foster et al. 2001). This finding is consistent with Jovanovic (1982)’s passive learning model, where a firm’s prior about its profitability are not terribly informative. Only entry provides information about profitability. The firm decides whether to grow, shrink, or exit by updating its best estimation of its productivity. Our model differs from Jovanovic’s in the source of the uncertainty about the firm’s profitability. We stress the quality of the match among owners. Data on owners demonstrates that experimentation and learning underlies frailty in young partnerships. The theoretical model builds on McAdams (2011), who formalizes the interaction of matched partners in a repeated prisoner’s dilemma. The option to incorporate, which we treat as an increase in dissolution costs, allows agents to choose stricter commitment. Earlier empirical studies address related questions. In his exploration of high-firm dissolution rates among young enterprises in the Argentinian and Irish newspaper industries, Levinthal (1991) argues that a sequence of initial successes can shield firms from failure later. We also find high mortality for young partnerships, but corporations in our dataset are relatively long-lived. Our evidence shows that organizations exhibit high rates of dissolution early on due to uncertainty regarding matches (and learning about the match quality through initial trial rounds), rather than the protection of initial successes against future shocks. The article proceeds as follows: Section 1 provides the historical background of Egypt’s legal institutions during the first half of the twentieth century. Section 2 describes this paper’s data. Section 3 develops a theory of partnerships by building on multi-armed bandit models. We illustrate the model’s implications using simulation, and then test these same implications empirically in Section 4. Section 5 summarizes the findings with concluding remarks. 1. Institutional Background Increasing European presence in 19th-century Egypt led to legal disarray. A concession system gave Europeans extraterritorial rights that included the right to contract under their own law and to litigate using their consular courts. This legal regime created uncertainty about just which law would apply to contractual relations between natives (whether Muslim or non-Muslim) and Europeans, or between Europeans of different nationalities.4 The Egyptian government attempted to rectify this problem in 1875 by introducing a new court system: the Mixed Courts. The new legal system operated independent of the executive. The Mixed Courts applied a slightly modified form of French law, followed French legal procedure, and held jurisdiction over all commercial and civil cases involving people of different national jurisdictions. The new courts also had jurisdiction over all Egyptians forming companies. The Mixed Courts operated in Alexandria, Cairo, and Mansoura and were responsible in each case for the city and surrounding area. Judges included Europeans as well as Egyptians. Both contemporaries and more recent observers view the Mixed Courts as impartial and independent.5 The government abolished the Mixed Courts in 1949 as part of a larger plan to nationalize the legal system. Both locals and Europeans relied on the Mixed Courts for their commercial affairs.6 The Mixed Courts dealt with a range of commercial issues, including company law. Any multi-owner enterprise operating primarily in Egypt had to register with the Mixed Courts and fell under the Courts’ jurisdiction (Scott 1907). The dataset described below spans the entire population of multi-owner enterprises in Cairo, Alexandria, and Mansoura. Because the Mixed Courts applied the French commercial code, they introduced new enterprise forms to the Egyptian economy. Table 1 shows these forms, which included the société en nom collectif (general partnership), the société en commandite simple (limited partnership), the société en commandite par actions (limited partnership with tradable shares), and the société anonyme (corporation).General partnerships had two or more general partners, all of whom had unlimited liability. Any one of these partners could participate in company management. Limited partnerships consisted of at least one general partner and one “special” or limited partner. General partners had unlimited liability and management rights; limited partners were only liable up to the amount they invested and had no rights over the firm’s management. Both general and limited partnerships could contract flexibly on cash-flow and control rights, subject to these limitations. Limited partnerships with tradable shares resembled limited partnerships, except that the limited partners’ capital was divided into shares which could be traded on the stock market. Agents were free to choose any of these forms when they wrote a partnership contract. Forming a corporation, however, required the Egyptian government’s permission. Before the Mixed Courts’ introduction, European enterprise forms were only available to European subjects under consular jurisdiction. The native Muslim population, on the other hand, had been using Islamic partnerships until the transplantation of this French-style commercial code.7 Table 1. Menu of Organizational Forms Type Notes Société en nom collectif General partnership Two or more unlimitedly liable partners. Société en commandite simple Limited partnership One or more unlimitedly liable partners. One or more partners with limited liability. Special partners cannot participate in management. Société en commandite par actions Limited partnership by shares One or more unlimitedly liable partners. One or more partners with limited liability. Special partners can have a board to monitor the ordinary partners, otherwise do not participate in management. Special partners’ shares can be traded on the market. Société anonyme Corporation Seven or more members. All members have limited liability. Members’ shares are tradable. Requires special permission from the government. Single board of directors with at least three members. Type Notes Société en nom collectif General partnership Two or more unlimitedly liable partners. Société en commandite simple Limited partnership One or more unlimitedly liable partners. One or more partners with limited liability. Special partners cannot participate in management. Société en commandite par actions Limited partnership by shares One or more unlimitedly liable partners. One or more partners with limited liability. Special partners can have a board to monitor the ordinary partners, otherwise do not participate in management. Special partners’ shares can be traded on the market. Société anonyme Corporation Seven or more members. All members have limited liability. Members’ shares are tradable. Requires special permission from the government. Single board of directors with at least three members. Open in new tab Table 1. Menu of Organizational Forms Type Notes Société en nom collectif General partnership Two or more unlimitedly liable partners. Société en commandite simple Limited partnership One or more unlimitedly liable partners. One or more partners with limited liability. Special partners cannot participate in management. Société en commandite par actions Limited partnership by shares One or more unlimitedly liable partners. One or more partners with limited liability. Special partners can have a board to monitor the ordinary partners, otherwise do not participate in management. Special partners’ shares can be traded on the market. Société anonyme Corporation Seven or more members. All members have limited liability. Members’ shares are tradable. Requires special permission from the government. Single board of directors with at least three members. Type Notes Société en nom collectif General partnership Two or more unlimitedly liable partners. Société en commandite simple Limited partnership One or more unlimitedly liable partners. One or more partners with limited liability. Special partners cannot participate in management. Société en commandite par actions Limited partnership by shares One or more unlimitedly liable partners. One or more partners with limited liability. Special partners can have a board to monitor the ordinary partners, otherwise do not participate in management. Special partners’ shares can be traded on the market. Société anonyme Corporation Seven or more members. All members have limited liability. Members’ shares are tradable. Requires special permission from the government. Single board of directors with at least three members. Open in new tab 2. Data and Sources This article’s empirical results rely on a new, firm-level dataset based on the activities of the Mixed Courts. The dataset includes more than 12,000 firms with 28,000 individual partners. Our results on learning and match quality rests on this dataset’s most unusual feature: the detailed history of individual owners. The firm-level data ordinarily used in economics lacks this information. This section describes the sources and provides information necessary to the discussion here and in Section 4. Supplementary Appendix provides more information on the dataset. We construct our data from entries printed in an official serial. Starting in November 1910, Egyptian Mixed Courts published a monthly periodical in French, Gazette des Tribunaux mixtes d’Égypte. Another newspaper, Journal des Tribunaux mixtes began in November 1921. This second publication terminated on March 31, 1949, which marks the end of the dataset. The Gazette printed (among other things) extracts of partnership agreements, modifications to existing agreements, and dissolutions registered at the Mixed Courts. In November 1921, the Journal took over the publication of these notices.8 The law required partnerships to register their contracts at the local court and to publish summaries of these contracts in the newspapers. Companies risked annulment if they did not comply, so the dataset includes all multi-owner enterprises. Published extracts had to contain the following information: company name, enterprise form, general partners’ names, designation of partners who had authorization to manage or administer the company and sign for the firm, capital provided by shares or by limited partners, the firm’s start date, and, sometimes, a contracted duration.9 The end date specified in partnership contracts communicated little information. By law, partners always had the right to withdraw from the company (Egypt, 3 Décembre 1890, Bulletin de Législation et de Jurisprudence, III, 43). In addition, the end of the fixed term did not necessarily lead to the company’s dissolution. The law assumed that the company continued its operations past the contracted duration unless the partners registered an act of dissolution with the Mixed Courts (Egypt, 6 Avril 1893, Bulletin de Législation et de Jurisprudence, V, 291). In other words, the contracted duration could neither prevent partners unilaterally withdrawing from the firm nor trigger automatic dissolution at the end of the term. As such, many firms did not specify any duration at all; those that did so picked whole numbers (usually 2 or 3 years) that ended up overstating their realized duration. Our empirical tests require both the firm’s formation date and dissolution date. We study businesses established between October 1, 1910 and March 31, 1949. Firm histories can be censored on both sides. Some firms dissolve after the first date and we lack any record of their creation. Others dissolved after March 31, 1949, but we do not know when. We ignore firms created prior to October 1, 1910. To contend with firms that dissolved after our window, we apply a consistent censoring strategy described below. De-registration information is not as complete as other parts of the dataset. By law, parties had to register their firm’s dissolution. The law did not, however, specify how this rule was enforced. To overcome this problem, we rely on information about operating firms taken from the Egyptian Directory, which was published annually from 1907 well into the 1950s. These volumes provide a comprehensive list of all commercial enterprises operating in various cities in Egypt. By cross-checking every enterprise without a dissolution date against firms listed in the directories, we determined the last year a firm appeared to be in operation. If a firm appeared in the directory for 1930 but not 1931, for example, we conclude it wound-up operations in 1930. We assume that companies that had no de-registration dates but appeared in all directories up to and including the 1950 directory survived until beyond our window. This procedure is imperfect, as we do not know precisely when these enterprises ceased operations. Since our empirical analysis explores binary survival outcomes over periods of 1, 2, 5 years, etc., this imprecision matters little for our results.10 The dataset contains individual-level information about firm owners: general partners in partnerships and founding subscribers in corporations. We use owners’ names to code ethno-religious identifiers for each individual (e.g., Muslim, Jewish, Greek). We also matched partners across multiple partnerships using their names, cross-checking with the information in the commercial directory. Using each individual’s last name, or their fathers’ names (usually specified as part of the individual’s name), we also identify family firms. We consider an enterprise a family firm if more than four-fifths of all general partners were related in a general partnership, if more than half of shares were owned by members of the same family in a corporation, or if the firm called itself a family company by using an identifier such as “sons,” “brothers,” or “cousins.” This procedure yields a lower-bound count of family firms, because we cannot identify son-in-laws or married sisters. Few enterprises list limited partners by name, so we also cannot identify some family connections of this type.11 This unusual information on owner name and identity allows us to test our hypotheses concerning learning about match quality. 2.1 An Overview of Egyptian Enterprises The new dataset allows us to establish five new facts that motivate modeling choices in the theory and empirical sections. First, the learning we emphasis occurred against a backdrop of considerable churning. Many firms entered and exited every year, as Figure 1 shows.12 Periods of economic upturn coincided with a spike in the formation and dissolution of new companies. Similarly, fewer firms formed or dissolved during economic downturns. Artunç (2018) examines this relationship econometrically and shows that firm cohorts established during recessions were different from those established during expansions in many dimensions, including entrant rates, capitalization, and number of partners. Figure 1. Open in new tabDownload slide Formation and Dissolution, 1911–1948. Source: See text for firm formation and dissolution. For U.S. cotton price, see Olmstead and Rhode (2016). Figure 1. Open in new tabDownload slide Formation and Dissolution, 1911–1948. Source: See text for firm formation and dissolution. For U.S. cotton price, see Olmstead and Rhode (2016). Second, the partnership was by far the most common enterprise form. Figure 2 plots the distribution of companies across enterprise forms between 1916 and 1945. Just as in Europe, most Egyptian firms organized as general or limited partnerships.13 We see little systematic evolution in the choice of enterprise form over time. Overall, 53% of all new firms formed as general partnerships and 41% as limited partnerships. Not surprisingly, corporations were less common than general or limited partnerships. This difference reflects the costly process of acquiring permission to incorporate from the government as well as our larger point, that entrepreneurs only formed corporations when they wanted to commit to long-term business plans. Egypt indeed placed unusual restrictions on access to the corporation, but that form’s prevalence differed little from countries such as France and Germany that did not require concessions.14 Figure 2. Open in new tabDownload slide New Multi-owner Firms by Enterprise Form, 1916–1945. Figure 2. Open in new tabDownload slide New Multi-owner Firms by Enterprise Form, 1916–1945. Third, most companies had only two partners and small capitalization. We have two ways to measure firm size: the number of partners and the initial capitalization. Partner numbers and capitalization affect the firm’s chance of survival. First, any single general partner can force dissolution in a partnership. We might expect the risk of dissolution to increase with the number of general partners.15 Second, some partnerships might break up early because of low capitalization. Small firms might be more vulnerable to negative shocks than large companies. Third, on a related point, one might think that the firm survival rates reflect the concentration of enterprise forms in specific industries. Table 2 considers size. Most firms were small, with two general partners on average, although there were some larger firms at the upper tail. Table 3 unpacks these averages by reporting the size distribution by different legal forms. Seventy-one per cent of all general partnerships had two general partners, most of whom had joint control and management. Thus, the modal general partnership consisted of two agents who, as a matter of default rules, shared control rights over the company. In comparison, most limited partnerships involved one general partner and one limited partner.16 Table 2. Multi-Owned Enterprises: Summary Statistics General Limited Ltd partnership Corporation partnership partnership with shares Court of registration (column %) Alexandria 45.26 51.67 45.71 50.12 Cairo 49.95 45.23 52.38 48.71 Mansoura 4.78 3.09 1.90 1.17 Number of partners General 2.40 1.44 1.58 partners (0.79) (0.70) (1.60) [2, 14] [1, 9] [1, 23] Special 1.46 partners (1.10) [1, 37] Firm capital (100,000s of 2010 £ ⁠) Overall 2.79 3.47 13.98 37.64 (14.39) (17.93) (41.88) (76.93) [0.0105, 544.10] [0.0315, 788.31] [0.178, 500.97] [0.0054, 655.79] Per general 1.15 2.48 partner (6.20) (9.63) [0.0052, 272.05] [0.0169, 394.15] Partner characteristics (column %) 1+ Muslim 14.39 9.33 19.37 64.95 Mixed 29.46 12.83 18.73 86.45 Family 28.31 12.53 11.11 11.21 Family-2 28.31 27.66 25.93 Industry (column %) Wholesale/retail 52.38 53.13 33.97 17.99 Manufacturing 16.25 15.43 31.43 30.37 Finance 1.77 2.45 7.62 8.41 Other 29.6 28.99 26.98 43.22 General Limited Ltd partnership Corporation partnership partnership with shares Court of registration (column %) Alexandria 45.26 51.67 45.71 50.12 Cairo 49.95 45.23 52.38 48.71 Mansoura 4.78 3.09 1.90 1.17 Number of partners General 2.40 1.44 1.58 partners (0.79) (0.70) (1.60) [2, 14] [1, 9] [1, 23] Special 1.46 partners (1.10) [1, 37] Firm capital (100,000s of 2010 £ ⁠) Overall 2.79 3.47 13.98 37.64 (14.39) (17.93) (41.88) (76.93) [0.0105, 544.10] [0.0315, 788.31] [0.178, 500.97] [0.0054, 655.79] Per general 1.15 2.48 partner (6.20) (9.63) [0.0052, 272.05] [0.0169, 394.15] Partner characteristics (column %) 1+ Muslim 14.39 9.33 19.37 64.95 Mixed 29.46 12.83 18.73 86.45 Family 28.31 12.53 11.11 11.21 Family-2 28.31 27.66 25.93 Industry (column %) Wholesale/retail 52.38 53.13 33.97 17.99 Manufacturing 16.25 15.43 31.43 30.37 Finance 1.77 2.45 7.62 8.41 Other 29.6 28.99 26.98 43.22 Notes: Where applicable, standard deviations are reported in parentheses, minimums and maximums are reported in brackets. The variable “Muslim” indicates whether the partnership has at least one Muslim partner. “Mixed” denotes companies in which at least two general partners (for partnerships) or founders (for corporations) have different ethno-religious backgrounds. “Family-2” restricts the sample to partnerships with at least two general partners (matters only for limited partnerships and limited partnerships with tradeable shares). The overall sample consists of 6,398 general partnerships, 4,815 limited partnerships, 315 limited partnerships with tradable shares, and 427 corporations. Initial capitalization was reported for 2,863 general partnerships, 2,945 limited partnerships, 299 limited partnerships with tradable shares, and 424 corporations. Open in new tab Table 2. Multi-Owned Enterprises: Summary Statistics General Limited Ltd partnership Corporation partnership partnership with shares Court of registration (column %) Alexandria 45.26 51.67 45.71 50.12 Cairo 49.95 45.23 52.38 48.71 Mansoura 4.78 3.09 1.90 1.17 Number of partners General 2.40 1.44 1.58 partners (0.79) (0.70) (1.60) [2, 14] [1, 9] [1, 23] Special 1.46 partners (1.10) [1, 37] Firm capital (100,000s of 2010 £ ⁠) Overall 2.79 3.47 13.98 37.64 (14.39) (17.93) (41.88) (76.93) [0.0105, 544.10] [0.0315, 788.31] [0.178, 500.97] [0.0054, 655.79] Per general 1.15 2.48 partner (6.20) (9.63) [0.0052, 272.05] [0.0169, 394.15] Partner characteristics (column %) 1+ Muslim 14.39 9.33 19.37 64.95 Mixed 29.46 12.83 18.73 86.45 Family 28.31 12.53 11.11 11.21 Family-2 28.31 27.66 25.93 Industry (column %) Wholesale/retail 52.38 53.13 33.97 17.99 Manufacturing 16.25 15.43 31.43 30.37 Finance 1.77 2.45 7.62 8.41 Other 29.6 28.99 26.98 43.22 General Limited Ltd partnership Corporation partnership partnership with shares Court of registration (column %) Alexandria 45.26 51.67 45.71 50.12 Cairo 49.95 45.23 52.38 48.71 Mansoura 4.78 3.09 1.90 1.17 Number of partners General 2.40 1.44 1.58 partners (0.79) (0.70) (1.60) [2, 14] [1, 9] [1, 23] Special 1.46 partners (1.10) [1, 37] Firm capital (100,000s of 2010 £ ⁠) Overall 2.79 3.47 13.98 37.64 (14.39) (17.93) (41.88) (76.93) [0.0105, 544.10] [0.0315, 788.31] [0.178, 500.97] [0.0054, 655.79] Per general 1.15 2.48 partner (6.20) (9.63) [0.0052, 272.05] [0.0169, 394.15] Partner characteristics (column %) 1+ Muslim 14.39 9.33 19.37 64.95 Mixed 29.46 12.83 18.73 86.45 Family 28.31 12.53 11.11 11.21 Family-2 28.31 27.66 25.93 Industry (column %) Wholesale/retail 52.38 53.13 33.97 17.99 Manufacturing 16.25 15.43 31.43 30.37 Finance 1.77 2.45 7.62 8.41 Other 29.6 28.99 26.98 43.22 Notes: Where applicable, standard deviations are reported in parentheses, minimums and maximums are reported in brackets. The variable “Muslim” indicates whether the partnership has at least one Muslim partner. “Mixed” denotes companies in which at least two general partners (for partnerships) or founders (for corporations) have different ethno-religious backgrounds. “Family-2” restricts the sample to partnerships with at least two general partners (matters only for limited partnerships and limited partnerships with tradeable shares). The overall sample consists of 6,398 general partnerships, 4,815 limited partnerships, 315 limited partnerships with tradable shares, and 427 corporations. Initial capitalization was reported for 2,863 general partnerships, 2,945 limited partnerships, 299 limited partnerships with tradable shares, and 424 corporations. Open in new tab Table 3. Partnership Size Number of firms Number of general partners General partnerships Limited partnerships Limited with shares 1 3,129 207 (65.0) (65.7) 2 4,596 1,336 79 (71.8) (27.8) (25.1) 3 1,303 277 19 (20.4) (5.8) (6.0) 4 344 54 3 (5.4) (1.1) (1.0) 5 or more 156 19 7 (2.4) (0.4) (2.2) Total 6,399 4,815 315 Number of firms Number of general partners General partnerships Limited partnerships Limited with shares 1 3,129 207 (65.0) (65.7) 2 4,596 1,336 79 (71.8) (27.8) (25.1) 3 1,303 277 19 (20.4) (5.8) (6.0) 4 344 54 3 (5.4) (1.1) (1.0) 5 or more 156 19 7 (2.4) (0.4) (2.2) Total 6,399 4,815 315 Notes: Column percentages are reported in parentheses. Open in new tab Table 3. Partnership Size Number of firms Number of general partners General partnerships Limited partnerships Limited with shares 1 3,129 207 (65.0) (65.7) 2 4,596 1,336 79 (71.8) (27.8) (25.1) 3 1,303 277 19 (20.4) (5.8) (6.0) 4 344 54 3 (5.4) (1.1) (1.0) 5 or more 156 19 7 (2.4) (0.4) (2.2) Total 6,399 4,815 315 Number of firms Number of general partners General partnerships Limited partnerships Limited with shares 1 3,129 207 (65.0) (65.7) 2 4,596 1,336 79 (71.8) (27.8) (25.1) 3 1,303 277 19 (20.4) (5.8) (6.0) 4 344 54 3 (5.4) (1.1) (1.0) 5 or more 156 19 7 (2.4) (0.4) (2.2) Total 6,399 4,815 315 Notes: Column percentages are reported in parentheses. Open in new tab The capitalization data requires caution because of possible sample selection issues, but one can infer patterns from what is available. Table 2 shows that partnerships on average had much smaller capitalization than did corporations. Variation in capitalization within an enterprise form was also considerable. Figure 3 shows the capitalization distribution of multi-owned enterprises by legal form, expressed in the natural logarithm of pounds sterling (2010 prices). A wide range of firms, from small grocery stores to large land companies, preferred to organize their business as partnerships. The two largest firms in the dataset organized as a general and a limited partnership. The two partnership forms had a higher concentration of firms in the lower tail of capitalization, however. The largest firms overall organized as corporations, followed by share partnerships. Figure 4 reports the weight of corporations in numbers and in total capital invested, by year. Corporations were a small but growing share of all firms; because of their much larger size, they always accounted for at least 40% of all capital invested in multi-owner firms. Figure 3. Open in new tabDownload slide Capital Distribution by Enterprise Form. Notes: Capital is expressed as the logarithm of pounds sterling in 2010 prices. The upper horizontal axis gives the corresponding pounds sterling equivalent without taking logs. Figure 3. Open in new tabDownload slide Capital Distribution by Enterprise Form. Notes: Capital is expressed as the logarithm of pounds sterling in 2010 prices. The upper horizontal axis gives the corresponding pounds sterling equivalent without taking logs. Figure 4. Open in new tabDownload slide The Share of Corporations in Incumbent Firms and Firm Capital by Year. Notes: We estimate the share of corporate capital in total capital of all incumbent firms of our dataset in each year under three different assumptions. The “naive” method assumes that partnerships that did not report capital had no capital at all. The “upper bound” method assumes that the average capital invested in firms that did not report their capitalization corresponded to the 25th percentile of the general and limited partnerships that did report. The “lower bound” method assumes that the average capital of these partnerships with no capitalization data was equal to the 75th percentile of the general and limited partnerships with capitalization data. Figure 4. Open in new tabDownload slide The Share of Corporations in Incumbent Firms and Firm Capital by Year. Notes: We estimate the share of corporate capital in total capital of all incumbent firms of our dataset in each year under three different assumptions. The “naive” method assumes that partnerships that did not report capital had no capital at all. The “upper bound” method assumes that the average capital invested in firms that did not report their capitalization corresponded to the 25th percentile of the general and limited partnerships that did report. The “lower bound” method assumes that the average capital of these partnerships with no capitalization data was equal to the 75th percentile of the general and limited partnerships with capitalization data. Fourth, ethnically heterogeneous partnerships were rare, and family firms accounted for a substantial proportion of enterprises. Table 2 also summarizes these findings. The table also shows that commercial companies were more common as partnerships than as corporations, but there were, in fact, some commercial corporations. The choice of enterprise forms differs little across sector. Three exceptions, cotton manufacturing, finance, and land companies, did attract a higher share of corporations, likely due to the larger capital needs of these industries. Sole proprietorships did not register and thus do not appear in our data. This has two implications for our results. First, there is a different margin, one between the sole-owner and multi-owner enterprise that we cannot study. Our model and results pertain to the contracting issues in multi-owner firms. Second, we ordinarily do not know whether a partner in one of our multi-owner firms has prior business experience as a sole proprietor. This implies some imprecision in our measure of prior business experience; “prior business experience” here really means prior experience in a multi-owner firm. Since sole proprietors are often small and informal, the few datasets that include ownership information usually exclude such enterprises. 3. A Model of Partnerships as Experimentation This section develops a model based on a multi-armed bandit framework to better understand the dynamics of partnership formation and dissolution.17 Agents match anonymously. Each pair is an “arm” that produces a flow return every period. We introduce two innovations. The choice of legal form pins down dissolution costs, and we endogenize dissolution costs by allowing partners to choose their legal form. The learning mechanism described by the multi-armed bandit setting explains the otherwise puzzling high attrition rate for young partnerships. An alternative account—that dissolutions reflect idiosyncratic shocks—might account for the higher dissolution rate for partnerships, but not the age pattern of that risk. This insight demonstrates a trade-off: low dissolution costs aid learning and experimentation, allowing agents to try out several partners and sort into more productive matches. But they can limit investment in companies that are good matches by reducing incentives for within-firm cooperation. 3.1 The Benchmark Model We begin with a benchmark setting with no investment. The economy has a continuum of identical agents who form pairs to undertake a joint venture. Each pair produces a surplus that depends on partners’ complementarity θ, which is drawn independently and identically from a distribution F with full support on the unit interval. Neither partner is informed about the true value of θ, but the distribution function F is common knowledge. There are infinitely many discrete periods. Let t index the firm’s age. A partnership’s flow of rewards follows an i.i.d. Bernoulli process; each partner receives reward xt = 1 with probability θ and nothing otherwise. Agents share a common discount factor δ∈(0,1) over time. The timing in each period is as follows: Each agent in an existing partnership decides to dissolve the firm at cost d > 0 or no. Agents without a partner are matched one-to-one. For each new match, θ is drawn. Each partner receives payoff xt. The equilibrium concept is Markov Perfect in symmetric strategies. Let π0 denote the common prior belief on θ. The state variable in each period is the posterior belief πt on θ. πt describes agents’ best prediction of θ based on the initial distribution F and the string of successes and failures they have observed. Agents use Bayes’ rule to update the posterior beliefs on the distribution in each period the partnership is active. πt is a sufficient statistic that summarizes the information agents share after t trials. Strategies are defined as σ:[0,1]→△({0,1}) ⁠, a mapping from posterior expectations into a mixture over leaving the partnership 0, or doing one more trial, 1. The benchmark setting describes a multi-armed bandit model with infinitely many arms with no recall and a one-time switching cost. Bergemann and Välimäki (2001) prove that, in this setting, the optimal policy is the Gittins index rule. The Gittins index describes the constant stream of rewards for which an agent is indifferent between doing one more trial and stopping to receive that constant reward forever. The index rule is to experiment with the arm that has the highest Gittins index. Since all potential matches have the same Gittins index, partners will continue experimenting in the current match until they receive a sufficiently high number of failures over successes so that the Gittins index of the active match drops below the Gittins index of potential matches minus the cost of dissolution. The model implies considerable churning, and that churning is optimal. If partners could not dissolve these associations at will, they might become stuck in unproductive ventures for some periods of time. The unproductive ventures would depress overall surplus in the economy due to the opportunity cost of unrealized but more productive matches. As the dissolution cost d increases, partnerships become more stable, as each matched pair must observe relatively more failures before dissolving the partnership. High dissolution costs lock the parties into associations that they would have optimally dissolved otherwise, reducing the chances of a more complementary match. 3.2 Inducing Investment In the benchmark model, all dissolutions are timely and beneficial. Because commitment is worthless, the corporation is a deadweight loss. We now enrich the model by introducing strategic actions in the stage game, providing a clear trade-off between ease of dissolution and commitment. Suppose that in each period, agents can decide whether to undertake a costly investment. Investment decisions are non-cooperative and the investing partner alone bears the cost c > 0. If neither agent undertakes the investment, each receives a flow payoff of xt. If only one agent invests, the partnership produces the flow rx for both, but that investor pays a cost of c. If both agents invest, each receives Rx – c. The flow variable x follows the Bernoulli process depending on the unobserved match quality θ. We assume c/2<R−r<c and R−1>c so that joint investment is socially efficient for partnerships with high enough θ but might not be sustained outside of a repeated game. Later on, we show that with sufficiently small dissolution cost, match-specific punishment mechanisms cannot induce cooperation since the defecting party can just take his outside option. We assume that partners observe investment decisions (hence, future plays can be conditioned on the history of investment decisions), but are unverifiable and thus cannot be enforced by a third party.18 Throughout the rest of this analysis, we focus on sub-game perfect equilibria in pure, symmetric strategies. Markov Perfection in this setting has no meaningful effect since the payoff-relevant state, namely the posterior belief π, is insufficient to sustain cooperation as an equilibrium.19 At the cost of creating a multitude of equilibria, sub-game perfection allows history-dependent strategies that can facilitate cooperation in equilibrium. We focus on the sub-game perfect equilibria that can most easily sustain investment, which are also the ex ante welfare maximizing equilibria. The extension transforms the model into a multi-armed bandit super-process. An agent is not merely experimenting with each “arm” but also considering an action that affects the arm’s payoffs. Whittle (1980) shows that the Gittins index policy is usually not optimal for multi-armed bandit super-processes unless the dominant action in each arm is independent of the constant stream of rewards to which one compares each arm’s continuation payoff. The independence condition fails in our setting, however, as the optimal decision to invest depends on the state variable πt and the outside option. Regardless, the equilibrium features a similar cutoff strategy. We use a simplified version with two types to illustrate the fundamental insights. The Supplementary Appendix presents a generalized model with continuous types. For now, suppose θ∈{L,H} where agents share a common prior that θ=H with probability π0. Matches of high quality H produce a positive surplus with probability q. Low matches, those with θ=L ⁠, never produce a positive surplus. Thus, as long as partners observe a failure, beliefs π decay according to Bayes’ rule: π′=π(1−q)/(1−πq) ⁠. If there is a success, then agents know the match quality to be H with certainty. Suppose there is no investment action at all. Then the setting reduces to a simpler version of the previous environment. Let V(π) denote the continuation payoff of an agent in equilibrium conditional on belief π. The continuation payoff of each agent can be written as V(π)=max{V(π0)−d,πq+πqδV(1)+(1−πq)V(π(1−q)1−πq)} (1) where V(1)=q/(1−δ) ⁠. This is essentially an optimal stopping problem. Dynamic programming reveals a cutoff π¯d below which agents dissolve the partnership and re-match. The critical π¯d can be recovered by setting V(π¯d)=V(π¯d(1−q)/(1−π¯dq))=V(π0)−d and solving for π¯d ⁠: π¯d=(1−δ)(V(π0)−d)q[1+δ(V(1)−V(π0)+d)]. (2) Both V(1) and V(π0) remain constant across all matches regardless of their current posteriors. In the extreme case of d = 0, agents immediately dissolve the firm if their first observation is a failure. Suppose now that agents can undertake an investment as described above. Assume that π0q(R−1)−c<0 ⁠, that is, investment is not efficient for any match if the match has no information in addition to the common prior. Sufficiently low dissolution costs with the re-matching option deters partners from investing even after they observe a success. Again, start with the extreme case of d = 0. If agents invest on the equilibrium path when π = 1, then V(1)=(Rq−c)/(1−δ) and V(π0)=π0q+π0qδ(Rq−c1−δ)+(1−π0q)δV(π0) (3) which we can solve to find V(π0)=π0q[1−δ(1−Rq+c)](1−δ)[1−δ(1−π0q)]. (4) Note that investment actions following the realization of a success raise the value of both the active match and possible re-matching. Defection from investment can only be deterred if the stage-game benefit followed by the continuation payoff of taking re-matching is less than the continuation payoff of complying with investment. That is, an equilibrium in which both agents invest requires rq+δ{π0q[1−δ(1−Rq+c)](1−δ)[1−δ(1−π0q)]}≥Rq−c1−δ. (5) This condition cannot be satisfied if c or π0 are sufficiently large relative to q(R−r) ⁠. If 1<π0(1+c−q(R−r)) ⁠, then the set of discount factors δ that can support investment in a model with re-matching is smaller than the one without that option. If the likelihood of good matches increases, dissolving the firm becomes more attractive. Since each new match is anonymous, all partnership-specific history is lost in the new match. Thus, a defector cannot be punished in any re-match. Some friction is necessary to encourage investment, either through a positive cost of switching or through decreasing the likelihood of getting a good re-match. As the cost of dissolution increases, the constraint that the outside option of re-matching imposes becomes slacker. Suppose that investment cannot be induced but partners have the option to incorporate. Incorporation in this setting amounts to irrevocable commitment: it raises the dissolution cost arbitrarily high. In doing so, incorporation removes the outside option of dissolving and re-matching, thus giving within-partnership punishments more bite in deterring deviations from cooperation. New matches are not willing to incorporate. At this point, the match quality is uncertain. Agents prefer to observe a few trials and have the option to quit so that they can sort into matches with better quality. Once they observe a success, all uncertainty is removed. In this case, partners prefer to pay a one-time sunk cost κ, less than q(R−1)/(1−δ) ⁠, to transform their enterprise into a corporation. Hence, the model has an equilibrium of the following type: partners shirk so long as there are no successes, they quit after a critical number of failures, but they incorporate and invest after observing a success. The Supplementary Material Appendix presents a general version of the model with continuous types and obtains a more nuanced result that describes firm dynamics and choice of form across different matches. We show that, if the return from investment R differs across matches due to match quality, then there will be some matched pairs that can sustain investment without incorporating in equilibrium. The return from cooperation in that case is so high that defecting and then re-matching ceases to be credible. These firms receive no further benefit from the corporate form since the stage payoffs are high enough to stave off any deviation. In equilibrium, firms go through an “exploration” phase in which partners shirk from investment and learn about match quality. After enough observations, bad matches dissolve, intermediate matches incorporate and good matches remain as partnerships. In this environment, an exogenous, higher dissolution cost lowers ex ante welfare. High dissolution costs lead to investment sooner in all matches. However, the higher cost discourages experimentation with new partners. This is the negative effect on sorting. If agents can endogenously choose to shut down their outside option in cases where they wouldn’t have dissolved the partnership anyway, they can achieve higher continuation payoffs than before. Increasing the dissolution cost endogenously has two effects on other equilibrium behavior. The option to incorporate further not only raises the value of each potential re-match but also raises the value of experimentation in the current match. Depending on which effect dominates, the dissolution cutoff can rise or fall when the option to incorporate becomes available. The model thus has several predictions critical for enterprise formation, dissolution, and choice of form. First, agents prefer to organize their enterprises as partnerships absent any additional signal about quality. Second, these partnerships display a single peak in their dissolution risk, concentrated within the first few “years” of formation. Third, the surviving partnerships are stable and undertake more investment in the firm. Fourth, partnerships whose “true” match quality is better than potential matches, but not high enough to sustain cooperation, will find it profitable to incorporate in order to commit partners to investment. 3.3 Model Simulations To explore the model’s implications, we simulate partnership formations and dissolutions based on the benchmark model with no investment. We create 10,000 partners who are matched and re-matched, implying a stable stock of 5,000 partnerships at any given time. The time horizon is T = 240, where each period broadly represents one “month.” Match quality θ has a beta distribution with parameters α = 50 and β = 10. The common discount factor is δ=0.99 ⁠. We assume an exogenous dissolution cost of d = 0.035. These numbers best replicate the empirical evidence we present in Section 4. The general shape of the simulated risk profile remains robust to other parameter specifications. Each partner optimally dissolves their match if the Gittins index mt(αt,βt) associated with their association is lower than a potential partnership’s, m0(α,β)=m0(50,10) ⁠. Partners update their posterior beliefs about θ at the end of each period. After t trials, they dissolve the partnership if agents believe that their firm’s performance falls short of what they expect to receive in a new match.20 Figure 5(a) shows the simulated hazard rate. Firms dissolve mostly in the first two “years.” A partnership is “safe” immediately after its foundation. After a short trial period, agents have a better idea about the firm’s productivity and decide to dissolve the partnership if they are pessimistic about their enterprise’s performance. Conditional on surviving this experimentation, partnerships become much less likely to dissolve. The simulation exercise also illustrates two comparative static implications. Figure 5(b) compares hazard rates for two different assumed dissolution costs, d′=0.047>0.035=d. Higher dissolution costs make partnerships more stable overall, reducing and delaying the early dissolution peak. This stability is suboptimal; partners might have better potential matches in expectation but remain in their current match due to the higher cost of breaking up. Figure 5(c) shows the case where agents have observed two “successes” prior to the game’s start and thus believe their match has high quality. These initial matches are more stable, in this case, optimally so. This comparison is especially relevant for family firms and partnerships whose partners had been in business together before. These matches have stronger signals on their joint productivity and the partnership creates less informational content for these agents. Figure 5. Open in new tabDownload slide Simulated Hazard Functions from the Model. Notes: The simulation assumes 10,000 agents who are matched over 240 periods, interpreting each period as a month. The figure uses 12-period intervals, hence each unit of time represents 1 year. The probability with which a partnership produces a positive reward is drawn from a beta distribution with parameters α = 50 and β = 10. The common discount factor is δ=0.99 ⁠. The baseline dissolution cost is d = 0.035, the high dissolution cost d′=0.047 ⁠. The prior information is equivalent to two observations of successes before start. See Section 3.3 for discussion. (a) Benchmark, (b) different dissolution costs, (c) prior information. Figure 5. Open in new tabDownload slide Simulated Hazard Functions from the Model. Notes: The simulation assumes 10,000 agents who are matched over 240 periods, interpreting each period as a month. The figure uses 12-period intervals, hence each unit of time represents 1 year. The probability with which a partnership produces a positive reward is drawn from a beta distribution with parameters α = 50 and β = 10. The common discount factor is δ=0.99 ⁠. The baseline dissolution cost is d = 0.035, the high dissolution cost d′=0.047 ⁠. The prior information is equivalent to two observations of successes before start. See Section 3.3 for discussion. (a) Benchmark, (b) different dissolution costs, (c) prior information. 4. Firm Survival in Egypt We now use the Egyptian data to estimate econometric models that test the core implication of the theoretical model. The linear probability models we report ask whether the firm survives a specific number of years. While we stress 5-year survival, the Supplementary Appendix reports parallel estimates for other durations. We are primarily interested in showing whether firms survive longer if their owners have characteristics that imply a better match from the outset: partners who have been in business together, for example. Rather than estimate a full structural model motivated by the theory, we consider reduced-form models that focus on the issue for which the data is best suited. Our approach requires that we treat as exogenous two variables that the theory treats as endogenous: corporation status and investment. This strategy represents a simplification made necessary by the data. The theory contemplates partners who match to form an enterprise, and who can later alter both their investment in the firm and its legal form. Since we only observe extant multi-owner firms we cannot capture the first-stage matching (which implies that the identity of partners is endogenous). The data include few firms that change investment or legal form, so any attempt to model the determinants of investment or legal form would rely primarily on cross-sectional variation in extant firms rather than transitions observed in the data. In our judgment, the compromises required by a structural approach are not worth the possibly restrictive assumptions needed. Abstracting from the endogeneity of enterprise form is, probably for this reason, a standard approach (Gregg 2015; Nicholas 2015). Instead, we proceed in two steps. We first estimate models of the determinants of corporation status. The Supplementary Appendix Table C2 reports linear probability models for which the dependent variable is one if the firm is a corporation, along with a parallel specification that treats the share partnership as identical to the corporation. These models used a regressor (“Title”) that equals one if any of the owners has the honorific title “Pasha,” “Efendi,” or “Bey.” These indicators capture whether an owner had the social status necessary to navigate the political process of securing a corporate concession. The variable has the expected sign and has the smaller magnitude we expect in the specification that includes share partnerships, which did not require the concession. Corporations have large capitalization, which is consistent with the fixed costs of incorporation. The other estimates have the signs predicted by the model. Corporations are less likely to be family firms because families do not need the commitment the corporation provides, and corporations are more likely than partnerships to have owners who know each other well. As the model implies, corporations are firms that have decided to commit because their owners have found a good match. We then turn to models of firm survival. These models treat corporate status as exogenous, which is a simplification. Many firms, however, did not have a realistic chance of incorporation: they lacked politically important partners. More importantly, the Supplementary Appendix Tables C14 and C15 report analogous models for ordinary and limited partnerships alone. The key results are nearly identical to those reported here. Figure 6(a) and 6(b) illustrates the underlying difference in frailty between partnerships and corporations for the first 10 years of their existence. The Kaplan-Meier survival functions in Figure 6(a) show that >10% of partnerships dissolved within the first year. The more stable corporations lasted much longer and their dissolution rates were constant over time. Figure 6(b) reports hazard rates for dissolution. Corporations had relatively steady death rates that did not vary significantly with firm age. A partnership’s conditional probability of dissolution, on the other hand, peaked around its first year of age; survivors enjoyed lower death rates that monotonically decreased in higher age groups, converging with corporations’ exit risk at the end of 10 years. Figure 6. Open in new tabDownload slide Survival and Hazard Estimates. Notes: The partnership category combines all three partnership forms (general, limited, and share). (a) Kaplan–Meier survival estimate. (b) Hazard rate estimate. Figure 6. Open in new tabDownload slide Survival and Hazard Estimates. Notes: The partnership category combines all three partnership forms (general, limited, and share). (a) Kaplan–Meier survival estimate. (b) Hazard rate estimate. The data have several observable characteristics that indicate whether partners knew each other when they created their firm. To understand how important these characteristics are, we need to hold constant other features such as firm size. We experimented with several approaches, including parametric failure-time models. These models require nontrivial functional-form assumptions we prefer to avoid. Here, we focus on more robust models that treat survival to a certain duration (such as 5 years) as a binary outcome. The Supplementary appendix presents alternative specifications, including robustness checks for the binary models explored here. The main results are robust to reasonable alternative approaches. Table 4 reports linear probability models in which the dependent variable takes a value of one when the firm survives to at least the specified age. We stress 5-year survival; the table also reports results for 2-year survival, and Tables C5–C7 in the Supplementary Appendix show results for other durations. Table 4 illustrates two unavoidable data issues. Many firms do not report their capitalization when they form. We estimate models with and without this regressor, and also estimate models constrained to the sub-sample lacking the capitalization data. The capitalization variable issue does not affect our primary result. The second data issue reflects the lack of de-registration information for some firms, noted in Section 3 above. We present results for two different treatments of this issue, but stress the specification that simply drops these firms, which amounts to assuming these firms never operated. Tables C10–C13 in the Supplementary Appendix present robustness checks for alternative assumptions. Table 4. Linear Probability Estimates of Firm Survival Estimates excluding firms with no survival information Incl firms with no survival info Dependent variable 60 mo 60 mo 60 mo 24 mo 24 mo 24 mo 60 mo 60 mo 60 mo Log capital 0.0379 0.0356 0.00243 (0.00602) (0.00494) (0.00550) Corporation 0.426 0.533 0.499 0.221 0.325 0.290 0.391 0.425 0.396 (0.0337) (0.0263) (0.0315) (0.0268) (0.0198) (0.0249) (0.0320) (0.0248) (0.0298) Limited partnership −0.0513 −0.0142 −0.0330 −0.0138 0.000351 0.00341 −0.0490 −0.0287 −0.0478 (0.0207) (0.0151) (0.0206) (0.0194) (0.0143) (0.0193) (0.0198) (0.0145) (0.0197) Share partnership 0.0256 0.117 0.0942 0.0257 0.102 0.0902 0.0231 0.0441 0.0277 (0.0424) (0.0375) (0.0411) (0.0344) (0.0302) (0.0334) (0.0397) (0.0351) (0.0383) Family firm 0.270 0.311 0.290 0.177 0.212 0.195 0.212 0.231 0.213 (0.0221) (0.0147) (0.0219) (0.0173) (0.0118) (0.0172) (0.0205) (0.0135) (0.0202) 2+ experienced owners −0.0565 −0.0563 −0.0400 −0.0540 −0.0384 −0.0386 −0.0808 −0.0926 −0.0796 (0.0285) (0.0211) (0.0284) (0.0265) (0.0206) (0.0265) (0.0278) (0.0208) (0.0276) 2+ in business together before 0.0833 0.0821 0.0928 0.0504 0.0597 0.0594 0.0637 0.0584 0.0641 (0.0342) (0.0277) (0.0342) (0.0309) (0.0261) (0.0308) (0.0336) (0.0274) (0.0336) Family×2+ before −0.0655 −0.0847 −0.0634 −0.0569 −0.0727 −0.0549 −0.0136 −0.0338 −0.0134 (0.0508) (0.0388) (0.0510) (0.0396) (0.0318) (0.0396) (0.0491) (0.0375) (0.0491) One general partner 0.119 0.0928 0.120 0.0765 0.0835 0.0770 0.114 0.0945 0.114 (0.0238) (0.0181) (0.0239) (0.0216) (0.0166) (0.0216) (0.0225) (0.0170) (0.0225) 3+ general partners 0.0190 0.0328 0.0292 0.00323 0.0103 0.0128 0.0303 0.0301 0.0310 (0.0220) (0.0147) (0.0220) (0.0195) (0.0133) (0.0195) (0.0204) (0.0137) (0.0203) Muslim/heterogeneous firm −0.0298 −0.0187 −0.0226 −0.0106 −0.00752 −0.00385 0.0231 0.0328 0.0235 (0.0289) (0.0225) (0.0286) (0.0262) (0.0210) (0.0260) (0.0265) (0.0207) (0.0265) Muslim/homogeneous firm 0.0527 0.0697 0.0734 0.00528 0.0351 0.0246 0.136 0.133 0.137 (0.0333) (0.0230) (0.0333) (0.0273) (0.0192) (0.0272) (0.0267) (0.0185) (0.0267) Non-Muslim/heterogeneous −0.0445 −0.0699 −0.0420 −0.0412 −0.0499 −0.0389 −0.0277 −0.0541 −0.0275 (0.0214) (0.0150) (0.0215) (0.0211) (0.0152) (0.0212) (0.0208) (0.0146) (0.0208) Constant −0.0899 0.214 0.332 0.233 0.491 0.629 0.508 0.302 0.534 (0.108) (0.0516) (0.0851) (0.116) (0.0499) (0.101) (0.106) (0.0501) (0.0871) Sub-sample limited to Firms with Full dataset Firms with Firms with Full dataset Firms with Firms with Full dataset Firms with K data K data K data K data K data K data Number of Observations 4,124 8,137 4,124 4,124 8,137 4,124 4,887 9,666 4,887 R2 0.154 0.129 0.146 0.083 0.076 0.073 0.092 0.084 0.092 Estimates excluding firms with no survival information Incl firms with no survival info Dependent variable 60 mo 60 mo 60 mo 24 mo 24 mo 24 mo 60 mo 60 mo 60 mo Log capital 0.0379 0.0356 0.00243 (0.00602) (0.00494) (0.00550) Corporation 0.426 0.533 0.499 0.221 0.325 0.290 0.391 0.425 0.396 (0.0337) (0.0263) (0.0315) (0.0268) (0.0198) (0.0249) (0.0320) (0.0248) (0.0298) Limited partnership −0.0513 −0.0142 −0.0330 −0.0138 0.000351 0.00341 −0.0490 −0.0287 −0.0478 (0.0207) (0.0151) (0.0206) (0.0194) (0.0143) (0.0193) (0.0198) (0.0145) (0.0197) Share partnership 0.0256 0.117 0.0942 0.0257 0.102 0.0902 0.0231 0.0441 0.0277 (0.0424) (0.0375) (0.0411) (0.0344) (0.0302) (0.0334) (0.0397) (0.0351) (0.0383) Family firm 0.270 0.311 0.290 0.177 0.212 0.195 0.212 0.231 0.213 (0.0221) (0.0147) (0.0219) (0.0173) (0.0118) (0.0172) (0.0205) (0.0135) (0.0202) 2+ experienced owners −0.0565 −0.0563 −0.0400 −0.0540 −0.0384 −0.0386 −0.0808 −0.0926 −0.0796 (0.0285) (0.0211) (0.0284) (0.0265) (0.0206) (0.0265) (0.0278) (0.0208) (0.0276) 2+ in business together before 0.0833 0.0821 0.0928 0.0504 0.0597 0.0594 0.0637 0.0584 0.0641 (0.0342) (0.0277) (0.0342) (0.0309) (0.0261) (0.0308) (0.0336) (0.0274) (0.0336) Family×2+ before −0.0655 −0.0847 −0.0634 −0.0569 −0.0727 −0.0549 −0.0136 −0.0338 −0.0134 (0.0508) (0.0388) (0.0510) (0.0396) (0.0318) (0.0396) (0.0491) (0.0375) (0.0491) One general partner 0.119 0.0928 0.120 0.0765 0.0835 0.0770 0.114 0.0945 0.114 (0.0238) (0.0181) (0.0239) (0.0216) (0.0166) (0.0216) (0.0225) (0.0170) (0.0225) 3+ general partners 0.0190 0.0328 0.0292 0.00323 0.0103 0.0128 0.0303 0.0301 0.0310 (0.0220) (0.0147) (0.0220) (0.0195) (0.0133) (0.0195) (0.0204) (0.0137) (0.0203) Muslim/heterogeneous firm −0.0298 −0.0187 −0.0226 −0.0106 −0.00752 −0.00385 0.0231 0.0328 0.0235 (0.0289) (0.0225) (0.0286) (0.0262) (0.0210) (0.0260) (0.0265) (0.0207) (0.0265) Muslim/homogeneous firm 0.0527 0.0697 0.0734 0.00528 0.0351 0.0246 0.136 0.133 0.137 (0.0333) (0.0230) (0.0333) (0.0273) (0.0192) (0.0272) (0.0267) (0.0185) (0.0267) Non-Muslim/heterogeneous −0.0445 −0.0699 −0.0420 −0.0412 −0.0499 −0.0389 −0.0277 −0.0541 −0.0275 (0.0214) (0.0150) (0.0215) (0.0211) (0.0152) (0.0212) (0.0208) (0.0146) (0.0208) Constant −0.0899 0.214 0.332 0.233 0.491 0.629 0.508 0.302 0.534 (0.108) (0.0516) (0.0851) (0.116) (0.0499) (0.101) (0.106) (0.0501) (0.0871) Sub-sample limited to Firms with Full dataset Firms with Firms with Full dataset Firms with Firms with Full dataset Firms with K data K data K data K data K data K data Number of Observations 4,124 8,137 4,124 4,124 8,137 4,124 4,887 9,666 4,887 R2 0.154 0.129 0.146 0.083 0.076 0.073 0.092 0.084 0.092 Notes: Robust standard errors in parentheses. All models include controls for sectors and foundation years. The Supplementary Appendix presents the full set of results for all coefficient estimates (Tables C2–C3 and C9), as well as robustness checks for different durations (12 months, 36 months, 120 months), different assumptions about firms for which there is no survival information, and restricted samples that only include certain sectors or legal forms. Open in new tab Table 4. Linear Probability Estimates of Firm Survival Estimates excluding firms with no survival information Incl firms with no survival info Dependent variable 60 mo 60 mo 60 mo 24 mo 24 mo 24 mo 60 mo 60 mo 60 mo Log capital 0.0379 0.0356 0.00243 (0.00602) (0.00494) (0.00550) Corporation 0.426 0.533 0.499 0.221 0.325 0.290 0.391 0.425 0.396 (0.0337) (0.0263) (0.0315) (0.0268) (0.0198) (0.0249) (0.0320) (0.0248) (0.0298) Limited partnership −0.0513 −0.0142 −0.0330 −0.0138 0.000351 0.00341 −0.0490 −0.0287 −0.0478 (0.0207) (0.0151) (0.0206) (0.0194) (0.0143) (0.0193) (0.0198) (0.0145) (0.0197) Share partnership 0.0256 0.117 0.0942 0.0257 0.102 0.0902 0.0231 0.0441 0.0277 (0.0424) (0.0375) (0.0411) (0.0344) (0.0302) (0.0334) (0.0397) (0.0351) (0.0383) Family firm 0.270 0.311 0.290 0.177 0.212 0.195 0.212 0.231 0.213 (0.0221) (0.0147) (0.0219) (0.0173) (0.0118) (0.0172) (0.0205) (0.0135) (0.0202) 2+ experienced owners −0.0565 −0.0563 −0.0400 −0.0540 −0.0384 −0.0386 −0.0808 −0.0926 −0.0796 (0.0285) (0.0211) (0.0284) (0.0265) (0.0206) (0.0265) (0.0278) (0.0208) (0.0276) 2+ in business together before 0.0833 0.0821 0.0928 0.0504 0.0597 0.0594 0.0637 0.0584 0.0641 (0.0342) (0.0277) (0.0342) (0.0309) (0.0261) (0.0308) (0.0336) (0.0274) (0.0336) Family×2+ before −0.0655 −0.0847 −0.0634 −0.0569 −0.0727 −0.0549 −0.0136 −0.0338 −0.0134 (0.0508) (0.0388) (0.0510) (0.0396) (0.0318) (0.0396) (0.0491) (0.0375) (0.0491) One general partner 0.119 0.0928 0.120 0.0765 0.0835 0.0770 0.114 0.0945 0.114 (0.0238) (0.0181) (0.0239) (0.0216) (0.0166) (0.0216) (0.0225) (0.0170) (0.0225) 3+ general partners 0.0190 0.0328 0.0292 0.00323 0.0103 0.0128 0.0303 0.0301 0.0310 (0.0220) (0.0147) (0.0220) (0.0195) (0.0133) (0.0195) (0.0204) (0.0137) (0.0203) Muslim/heterogeneous firm −0.0298 −0.0187 −0.0226 −0.0106 −0.00752 −0.00385 0.0231 0.0328 0.0235 (0.0289) (0.0225) (0.0286) (0.0262) (0.0210) (0.0260) (0.0265) (0.0207) (0.0265) Muslim/homogeneous firm 0.0527 0.0697 0.0734 0.00528 0.0351 0.0246 0.136 0.133 0.137 (0.0333) (0.0230) (0.0333) (0.0273) (0.0192) (0.0272) (0.0267) (0.0185) (0.0267) Non-Muslim/heterogeneous −0.0445 −0.0699 −0.0420 −0.0412 −0.0499 −0.0389 −0.0277 −0.0541 −0.0275 (0.0214) (0.0150) (0.0215) (0.0211) (0.0152) (0.0212) (0.0208) (0.0146) (0.0208) Constant −0.0899 0.214 0.332 0.233 0.491 0.629 0.508 0.302 0.534 (0.108) (0.0516) (0.0851) (0.116) (0.0499) (0.101) (0.106) (0.0501) (0.0871) Sub-sample limited to Firms with Full dataset Firms with Firms with Full dataset Firms with Firms with Full dataset Firms with K data K data K data K data K data K data Number of Observations 4,124 8,137 4,124 4,124 8,137 4,124 4,887 9,666 4,887 R2 0.154 0.129 0.146 0.083 0.076 0.073 0.092 0.084 0.092 Estimates excluding firms with no survival information Incl firms with no survival info Dependent variable 60 mo 60 mo 60 mo 24 mo 24 mo 24 mo 60 mo 60 mo 60 mo Log capital 0.0379 0.0356 0.00243 (0.00602) (0.00494) (0.00550) Corporation 0.426 0.533 0.499 0.221 0.325 0.290 0.391 0.425 0.396 (0.0337) (0.0263) (0.0315) (0.0268) (0.0198) (0.0249) (0.0320) (0.0248) (0.0298) Limited partnership −0.0513 −0.0142 −0.0330 −0.0138 0.000351 0.00341 −0.0490 −0.0287 −0.0478 (0.0207) (0.0151) (0.0206) (0.0194) (0.0143) (0.0193) (0.0198) (0.0145) (0.0197) Share partnership 0.0256 0.117 0.0942 0.0257 0.102 0.0902 0.0231 0.0441 0.0277 (0.0424) (0.0375) (0.0411) (0.0344) (0.0302) (0.0334) (0.0397) (0.0351) (0.0383) Family firm 0.270 0.311 0.290 0.177 0.212 0.195 0.212 0.231 0.213 (0.0221) (0.0147) (0.0219) (0.0173) (0.0118) (0.0172) (0.0205) (0.0135) (0.0202) 2+ experienced owners −0.0565 −0.0563 −0.0400 −0.0540 −0.0384 −0.0386 −0.0808 −0.0926 −0.0796 (0.0285) (0.0211) (0.0284) (0.0265) (0.0206) (0.0265) (0.0278) (0.0208) (0.0276) 2+ in business together before 0.0833 0.0821 0.0928 0.0504 0.0597 0.0594 0.0637 0.0584 0.0641 (0.0342) (0.0277) (0.0342) (0.0309) (0.0261) (0.0308) (0.0336) (0.0274) (0.0336) Family×2+ before −0.0655 −0.0847 −0.0634 −0.0569 −0.0727 −0.0549 −0.0136 −0.0338 −0.0134 (0.0508) (0.0388) (0.0510) (0.0396) (0.0318) (0.0396) (0.0491) (0.0375) (0.0491) One general partner 0.119 0.0928 0.120 0.0765 0.0835 0.0770 0.114 0.0945 0.114 (0.0238) (0.0181) (0.0239) (0.0216) (0.0166) (0.0216) (0.0225) (0.0170) (0.0225) 3+ general partners 0.0190 0.0328 0.0292 0.00323 0.0103 0.0128 0.0303 0.0301 0.0310 (0.0220) (0.0147) (0.0220) (0.0195) (0.0133) (0.0195) (0.0204) (0.0137) (0.0203) Muslim/heterogeneous firm −0.0298 −0.0187 −0.0226 −0.0106 −0.00752 −0.00385 0.0231 0.0328 0.0235 (0.0289) (0.0225) (0.0286) (0.0262) (0.0210) (0.0260) (0.0265) (0.0207) (0.0265) Muslim/homogeneous firm 0.0527 0.0697 0.0734 0.00528 0.0351 0.0246 0.136 0.133 0.137 (0.0333) (0.0230) (0.0333) (0.0273) (0.0192) (0.0272) (0.0267) (0.0185) (0.0267) Non-Muslim/heterogeneous −0.0445 −0.0699 −0.0420 −0.0412 −0.0499 −0.0389 −0.0277 −0.0541 −0.0275 (0.0214) (0.0150) (0.0215) (0.0211) (0.0152) (0.0212) (0.0208) (0.0146) (0.0208) Constant −0.0899 0.214 0.332 0.233 0.491 0.629 0.508 0.302 0.534 (0.108) (0.0516) (0.0851) (0.116) (0.0499) (0.101) (0.106) (0.0501) (0.0871) Sub-sample limited to Firms with Full dataset Firms with Firms with Full dataset Firms with Firms with Full dataset Firms with K data K data K data K data K data K data Number of Observations 4,124 8,137 4,124 4,124 8,137 4,124 4,887 9,666 4,887 R2 0.154 0.129 0.146 0.083 0.076 0.073 0.092 0.084 0.092 Notes: Robust standard errors in parentheses. All models include controls for sectors and foundation years. The Supplementary Appendix presents the full set of results for all coefficient estimates (Tables C2–C3 and C9), as well as robustness checks for different durations (12 months, 36 months, 120 months), different assumptions about firms for which there is no survival information, and restricted samples that only include certain sectors or legal forms. Open in new tab All specifications include controls for the firm’s sector and the year it formed. We include one dummy for firms that have a single general partner, and another for the small number of firms that have three or more. For corporations we code these variables as zero, so these dummies are implicitly interacted with a dummy for a form other than corporation. The reference firm for these variables is a two-person general partnership. The next seven regressors test the role of information and experimentation in the firm’s durability. We expect family firms to last longer, both because the owners know more about each other and because of implicit co-insurance commitments within the family. If a firm has two (or more) owners who have been in business together before, they should have more information about match quality and the firm will last longer. But someone who has been in business before just knows more about business; this is what Jovanovic (1982) captures with the firm’s priors about its profitability. We need to distinguish between what two owners know about each other from prior experience together on the one hand, and the effect of business experience in general, on the other. Thus, “in business together before” captures the pure effect of information about a specific pair. Because many family members form more than one business, we also include a control that interacts “in business together before” with the family-firm dummy. We include three more regressors that categorize the owners’ ethnicity. Some firms are homogeneous; their owners are all Muslim, for example, or all Greeks. Other firms have some Muslim and some non-Muslim owners. Finally, some firms have different kinds of non-Muslims (e.g., Greeks and Jews). The reference category for these variables is a firm consisting of a homogeneous group of unrelated non-Muslims (e.g., all Armenians) that has at most one partner with prior business experience. These variables do not, of course, fully account for the probability that a firm survives. The regression error term includes unobserved components of match quality, among other things. The possibility of omitted variables bias is always a concern in regression models, but for some omitted variable to affect our results, the omitted variable would have to be correlated with one of the included regressors. One can think of many reasons a firm would fail, but the plausible reasons would either be summarized by one of included variables (such as sector) or orthogonal to the included regressors (such as a bad business plan). Our results show that firms with more capital survive longer, as expected. But the second and third models reported in Table 4 do not differ appreciably in the estimates of interest. Being a corporation increases survival chances; share partnerships also last longer, although the effect is smaller. Having only one general partner also increases survival for limited partnerships. The sector dummies (not reported in Table 4, but in Supplementary Appendix Table C3) indicate that mines and other firms that would suffer heavily from untimely dissolution last longer. These effects, however, differ across the first three specifications.21 Family firms endured, as the theory implies. This result holds even when we consider the combined effect of being a family firm and having partners with business experience. A firm with two owners who have been in business together before also has a greater survival chance, although this effect is not always precisely estimated. When we also take into account that partners who have been in business together before are also more experienced business people, this family effect declines. Family members who have been in business before contribute less to the firm’s survival; depending on the specification, this point estimate completely offsets the effect of prior co-ownership. This result is to be expected, since family members acquire less (additional) information about one another from experimentation. The ethnicity results also demonstrate the role of learning about match quality. Members of the small Muslim business community knew each other, and had little reason to experiment; their firms endured. Business formed out of combinations of other groups had much learning to do, and the required experimentation led to firm break-up.22 Our results reflect two types of selection. Agents could choose different projects and with whom to partner. Selection into enterprise forms reflects these choices, and this selection is central to our argument. Agents choose the partnership form as long as their best guess of match quality is not high enough to justify commitment. The partnerships that last for 1 or 2 years belong to this group of experimenters. The surviving firms have much lower rates of breakdown. Being in a partnership does not “cause” a firm to dissolve quickly; the ease of dissolution is legally built into the form. The model and empirical analysis show that the selection of partners who are experimenting into partnerships drive the high frailty rates early on. Once the experimenters are gone, dissolution rates stabilize. Certain partner pairs such as family members or partners who have been in business before are not really experimenters. These firms did have more uniform dissolution rates. A second selection issue pertains to sectoral choices. Some firm activities, such as commodity brokerage or retail, do not require a lot of sunk capital to set up operations. Other industries involve high fixed and sunk capital investments. For example, most large-scale integrated cotton textile mills or steel mills would display such features. Naturally, these industries will not attract a lot of experimenters. Our results support this view. First, these industries constitute a higher share of corporations then they do of partnerships. Second, while experimentation is present in all sectors, it is strongest in mercantile companies and weakest in manufacturing firms. Checks reported in the Supplementary Appendix estimate separate regressions for several sectors. 5. Conclusion This article focuses on a neglected aspect of choices concerning the legal form of business enterprise: the ease of dissolution. We treat the partnership as a vehicle for experimentation that allows partners to match and learn about their match quality without making costly commitments to the enterprise. The modeling framework focuses on the trade-off between learning about match quality and the ability to commit to long-term projects. Because our data contains information on individual firm owners over (in many cases) several different firms, we can test the model’s lessons about match quality and prior knowledge about partners. The model and data have three important implications for firm longevity and enterprise form. First, general and limited partnerships indeed have shorter lives than corporations, even after we control for capitalization, sector, and other characteristics that might affect longevity. Most of the dissolution risk for partnerships, however, comes in the first 2 years after the firm’s establishment. Conditional on lasting 2 years, enterprise form matters less for surviving for five or more years. Second, the observed dissolution patterns support the model’s core idea: forming partnership reflects a process of experimentation with other investors whose ability cannot be observed perfectly. When two investors come from the same family, they have more prior information on each other, and their enterprise more likely endures. Similarly, when two individuals have been in business together before, they have a clearer idea of their suitability as partners in a business enterprise, and their firm lasts. These two effects partially substitute for one another; for family firms, the presence of a pair with prior joint experience matters less than for non-family firms. The extreme ethnic diversity of Egypt’s business community offers other clues to this effect. Firms created out of homogeneous partners endure, because these individuals knew each other. Other firms (perhaps created to capitalize on the gains from trade implicit in diversity) often do not last, because the partners went into the venture without knowing each other well. Third, this experimentation is beneficial. Consider the following counterfactual. If these forms could not be dissolved easily, then partners would have been stuck in unproductive ventures. Anticipating that result, agents would be less willing to enter into enterprises, depressing capital pooling and business activity as people waited for additional signals before establishing a multi-owner enterprise. Not all dissolutions, however, add to social welfare. Ease of dissolution depresses investment because partners cannot commit fully to the enterprise: some productive firms cannot survive without a way to lock-in the investments. While we have unusually rich data, some caveats are in order. Many firms lack capitalization data and for many there is no explicit de-registration event. Robustness checks reported in the text and Supplementary Appendix suggest these issues do not seriously bias the results, however. The variables we use to identify family firms and partners who have been in business together before are imperfect. In addition, our empirical results rest on a reduced-form approach that abstracts from some features of the theoretical model. More generally, the context we study may lend itself to more (or less) experimentation in ways that will only become clear with empirical studies in other times and places. The partnership form lies itself at the heart of the trade-off we identify. Rather than an unambiguously restrictive, inferior way of business organization, the partnership offers the best choice for agents who prefer not to lock capital in ventures that might turn out to be unproductive. It allows agents to trade off learning against larger investment. Some partnerships become suitable for incorporation after the trial period, while others will never incorporate. The partnerships that endure have such high match productivities that within-partnership punishments are still more valuable than the outside option of re-matching. Most partnerships that survived the initial trial period operated for decades without incorporating. Corporations, as economists, legal scholars, and historians have stressed, exist separately from their owners and thus provide an efficient vehicle for developing long-term investment projects, especially if those investments entail sunk costs. But those advantages, which amount to a greater ability to commit to long-term plans come at a cost: if other multi-owner forms did not exist, entrepreneurs would find themselves tied to investments without sufficient information on the project or the other owners. The ability to experiment with owners and projects without extensive commitment thus constitutes one of the partnership’s advantages. Acknowledgement This article is supported by the National Science Foundation under the grant NSF SES 1559273. We thank Naomi Lamoreaux, Jean-Laurent Rosenthal, Seven Ağir, Shameel Ahmad, Lint Barrage, Price Fishback, Amanda Gregg, Timur Kuran, Jakob Schneebacher, Gabriella Santangelo, Christopher Udry, and seminar participants at Harvard Business School, Yale University, and the World Economic History Congress 2015 for comments and suggestions. Laura D. Taylor provided excellent research assistance. The staff of Yale University Lillian Goldman Law Library, Bibliothèque nationale de France, and the British Library helped in locating sources. We also thank Roger Bilboul and David Lisbona for kindly sharing some of their private collection. An earlier draft of this article was circulated under the title “Enterprise Forms and Partnership Survival in Egypt between 1910 and 1949.” Funding National Science Foundation (NSF SES 1559273). Footnotes 1. A large majority of firms in our data are partnerships, which is typical of most countries prior to the introduction of private limited companies. The partnership form today, on the other hand, is rare. Levin and Tadelis (2005) note that partnerships in modern economies tend to be concentrated in sectors where employee human capital is central to the quality of the firm’s product and where customers cannot easily assess that human capital. While some of our partnerships may reflect these considerations, most are not in the sectors Levin and Tadelis (2005) stress: law and accounting firms, medical practices, etc. We find partnerships in sectors that have considerable physical capital and where the product is a physical object the quality of which the customer can observe. 2. Musacchio et al. (2008) study the longevity of partnerships in Mexico in the early 20th century. As part of a larger interest in the Mexican Revolution’s impact on the economy, they show that partnership life shortened considerably during the upheaval. 3. These models typically rely on multi-armed bandit frameworks. See Bergemann and Välimäki (2008) for an overview of bandit problems and their other applications in the literature. 4. Egypt was an autonomous province within the Ottoman Empire until the Empire’s dissolution after World War I. Artunç (2014) provides a comprehensive analysis of Ottoman legal pluralism. 5. Nationalists at the time denounced the Mixed Courts as tools of European domination. As Brown (1993) shows, these denunciations were unfounded; the Mixed Courts were established to limit foreign domination. Wilner (1975) agrees that the Mixed Courts limited privileges granted to foreigners. Hoyle (1991) goes further, showing that neither the British nor Egyptian government could interfere in the Courts’ work. The Mixed Courts judges were recruited from Egyptian Muslims and non-Muslims as well as Europeans; each group had to be represented to form a quorum (see Grigsby 1896; Herreros 1914). All judges were fluent in French and most were fluent in Arabic and English as well (Wilner 1975). Brinton, who sat on the Mixed Courts as a judge, praised the Mixed Courts as a judicial system as good as any of its European counterparts and for bringing order to Egypt’s judicial chaos (Brinton 1968). 6. Wood (2011: 46–51) shows that jurisdictional conflicts that had plagued the system earlier had been resolved by the early 1900s. Wilner (1975: 414) notes that “[the] cases brought before the Mixed Courts encompassed nearly all civil and commercial relations extant in the country.” 7. Native Christians and Jews could organize their firms using European partnership forms before the creation of the Mixed Courts by becoming protégés of European powers (Kuran 2004). There were, however, less than 2,000 protégés in the entire Ottoman Empire in the late 18th century (Artunç 2015: 727). The 1868 Egyptian census reveals that protégés accounted for less than 1% of non-Muslims in Egypt (Saleh 2018; Supplementary Appendix Table D6). 8. The two publications operated simultaneously after November 1921, but only the Journal published the notices used here after that date. 9. See Articles 54–58 of the Mixed Courts’ commercial code, Egypt (1907: 156–7). 10. We cross-checked this approach against approximately one-fifth of the direct reports of firm dissolution available and found substantial agreement. 11. While most firms reported their capitalization in nominal Egyptian pounds, others used pounds sterling, French Francs, or Ottoman pounds. We convert these values to pounds sterling using exchange rates cited in Denzel (2010), Owen (1993), and Pamuk (2000). These calculations use the historic opportunity cost method to express capitalization in real terms. We use pounds sterling as of 2010, http://www.measuringworth.com to adjust asset size in real terms. Historic opportunity cost is the most appropriate method for comparing firm capitalization. This method uses the GDP deflator to compare the cost index of all output in the economy. The British historic opportunity cost is suitable in this setting given Egypt’s close integration with the British economy and currency; see Yousef (2002). Many firms did not report their capitalization in the published extracts; we discuss this limitation in detail in Section 4 and in the Supplementary Appendix. 12. We omitted firms founded in 1910 and 1949 from this graph since the dataset starts in November 1910 and ends in March 1949. 13. Guinnane et al. (2007) show that ordinary partnerships accounted for 60% of all new companies in France even after the government introduced general incorporation in 1863. In Prussia (and later Germany), partnerships accounted for >80% of new firms until 1902 (pp. 702, 710–1). 14. The data contain a few private limited liability companies (PLLCs) registered in Alexandria and Cairo. The Mixed Courts did not recognize the PLLC as an enterprise form. The courts treated one of these PLLCs as being one of the three enterprise forms the code recognized; see Gazette des Tribunaux mixtes d’Égypte, v. 5, pp. 145–6; v. 12, pp. 17–8. This firm was founded in Germany as a GmbH. After 1929, more firms were registered as PLLCs, all of which were put under the jurisdiction of English law. Given their uncertain position, we exclude PLLCs from this analysis. 15. In societies such as Egypt in this period, mortality alone would dissolve many partnerships. A rough calculation from the UN model life tables “general, male” (http://www.un.org/esa/population/techcoop/DemMod/model_lifetabs/Model_LT_Annex3.pdf) provides an illustration. For a mortality level corresponding to an expectation of life at birth of 45 years, a considerable over-estimate for this time and place, there would be a 14% chance of at least one of two 40-year-old males dying within 5 years. With three partners at that age, the probability increases to 21%. For 60-year-old males, 31% of two-person and 43% of three-person partnerships would lose at least one owner to death in 5 years. These calculations treat the mortality risk of each partner as independent. 16. Although many partnerships specified the number of limited partners investing in the firm, many did not. The law itself does not mandate the disclosure of such information. Article 56 of the commercial code prohibits firms from reporting limited partners’ identity in the published extracts (Egypt 1907: 157). 17. See Bergemann and Välimäki (2008) for an overview of the framework’s applications. 18. McAdams (2011) proves the existence of a joint-welfare maximizing subgame perfect equilibrium in a more general repeated prisoner’s dilemma with heterogeneous agents and re-matching when partnership productivity is serially correlated. Our posterior belief π resembles his payoff-relevant random variable. But we make the learning component more explicit and focus on the dissolution costs. We also introduce an explicit choice to raise the cost of dissolution by deciding to organize the firm as a partnership or a corporation. 19. Markov Perfection is still viable if the timing is adjusted appropriately by allowing current investment decisions to affect future payoffs (in particular, by making stage payoffs exclusively depend on lagged investment decisions) or by introducing asynchronous investment actions to which agents can react. See Maskin and Tirole (1988) as an example. 20. Calculating the Gittins index is a complicated problem and there is no general analytical solution. However, there are many algorithms to approximate indices and solve value functions numerically. This simulation uses a closed form approximation. Whittle (1982) shows that, for large enough α+β and δ, the Gittins index of a Bernoulli sampling process is approximately m(α,β)≈αα+β+(αα+β)(1−αα+β)(α+β)(−2 log δ+1α+β)(αα+β)(1−αα+β)+αα+β−12. 21. In general, for models that include capitalization as a regressor, the sector dummies are not collectively significant. The sector effects are, on the other hand, significant for models that lack capitalization as a regressor. For the first model reported in Table 4, the F-statistic for the null hypothesis that all sector dummies are zero has a value of 1.26, which with degrees of freedom 10 and 4069 has a “p-value” of 0.2444. For the second model reported in that table, F(10,8083)=5.83 ⁠, and we can reject the null at conventional confidence level. The year-of-formation dummies, on the other hand, are always collectively significant; for the first two models reported in Table 4, F(29,4069)=4.12 for the first model and 3.94 for the second. This general pattern holds for all models reported in Table 4 and in the Supplementary Appendix. 22. We do not include the contracted duration as a regressor. As stated earlier, the law did not force any partner to be tied up in a firm if that partner decided to withdraw. Using this regressor would create a selection issue; about 15% of all partnerships did not specify a fixed term. Corporations had to specify a fixed period in their charters, but for such firms the duration figure represented their concession’s length. So the two concepts are not comparable. When partnerships did specify a fixed term, they bunched on whole numbers that overstated their realized duration; 26% picked 3 years and 24% 5 years. Fifty-six per cent of partnerships dissolved before their contracted termination date. Actual duration and contracted duration had a low correlation of 0.1225. As such, contracted durations give no real information. References Abramitzky Ran , Frank Zephyr , Mahajan Aprajit . 2010 . “Risk, Incentives, and Contracts: Partnerships in Rio De Janeiro, 1870–1891,” 70 Journal of Economic History 686 – 715 . Google Scholar Crossref Search ADS WorldCat Acemoglu Daron , Johnson Simon , Robinson James A. , 2001 . “The Colonial Origins of Comparative Development: An Empirical Investigation,” 91 American Economic Review 1369 – 1401 . Google Scholar Crossref Search ADS WorldCat Acemoglu Daron , Johnson Simon , Robinson James A. . 2002 . “Reversal of Fortune: Geography and Development in the Making of the Modern World Income Distribution,” 117 Quarterly Journal of Economics 1231 – 94 . Google Scholar Crossref Search ADS WorldCat Acemoglu Daron , Johnson Simon , Robinson James A. 2005 . “Institutions as a Fundamental Cause of Long-Run Growth,” 1 Handbook of Economic Growth 385 – 472 . Google Scholar Crossref Search ADS WorldCat Artunç Cihan. 2014 . “Barrators, Berats, and Bandits: Economic Implications of Legal Rules in the Ottoman Empire and Egypt, 1600–1921,” Ph.D. dissertation, Yale University. Artunç Cihan. 2015 . “The Price of Legal Institutions: The Beratl i Merchants in the Eighteenth-Century Ottoman Empire,” 75 Journal of Economic History 720 – 49 . Google Scholar Crossref Search ADS WorldCat Artunç Cihan. 2018 . “The Impact of Business Cycle Conditions on Firm Dynamics and Composition: Enterprise Entry and Exit in Egypt, 1911–48.” https://ssrn.com/abstract=3063818 or http://dx.doi.org/10.2139/ssrn.3063818. Banerjee Abhijit , Iyer Lakshmi . 2005 . “History, Institutions, and Economic Performance: The Legacy of Colonial Land Tenure Systems in India,” 95 American Economic Review 1190 – 1213 . Google Scholar Crossref Search ADS WorldCat Bartelsman Eric J. , Haltiwanger John , Scarpetta Stefano . 2004 . “Microeconomic Evidence of Creative Destruction in Industrial and Developing Countries,” IZA Discussion Paper No. 1374, Tinbergen Institute Discussion Papers No. TI 2004-114/3, World Bank Policy Research Paper No. 3464. Bergemann Dirk , Välimäki Juuso . 2001 . “Stationary Multi-choice Bandit Problems,” 25 Journal of Economic Dynamics and Control 1585 – 94 . Google Scholar Crossref Search ADS WorldCat Bergemann Dirk , Välimäki Juuso 2008 . “Bandit Problems,” in Durlauf Steven N. , Blume Larry E. , eds., The New Palgrave Dictionary of Economics . Basingstoke : Palgrave Macmillan . Google Preview WorldCat COPAC Blair Margaret M. , Stout Lynn A. . 2005 . “Specific Investment: Explaining Anomalies in Corporate Law,” 31 Journal of Corporation Law 719 . WorldCat Brinton Jasper Yeates. 1968 . The Mixed Courts of Egypt . New Haven : Yale University Press . Google Preview WorldCat COPAC Brown Nathan J. 1993 . “The Precarious Life and Slow Death of the Mixed Courts of Egypt,” 25 International Journal of Middle East Studies 33 – 52 . Google Scholar Crossref Search ADS WorldCat Chandler A. D. 1977 . The Visible Hand: The Managerial Revolution in American Business . Cambridge: Belknap Press . Google Preview WorldCat COPAC Cochran Thomas. 1977 . “The Business Revolution,” 79 American Historical Review 449 – 66 . WorldCat Dari-Mattiacci Giuseppe , Gelderblom Oscar , Jonker Joost , Perotti Enrico C. . 2017 . “The Emergence of the Corporate Form,” 33 Journal of Law, Economics, and Organization 193 – 236 . Google Scholar Crossref Search ADS WorldCat Decker Ryan , Haltiwanger John , Jarmin Ron , Miranda Javier . 2014 . “The Role of Entrepreneurship in us Job Creation and Economic Dynamism,” 28 Journal of Economic Perspectives 3 – 24 . Google Scholar Crossref Search ADS WorldCat Dell Melissa. 2010 . “The Persistent Effects of Peru’s Mining Mita,” 78 Econometrica 1863 – 1903 . Google Scholar Crossref Search ADS WorldCat Denzel Markus A. 2010 . Handbook of World Exchange Rates, 1590–1914 . Farnham: Ashgate . Google Preview WorldCat COPAC Dunne Timothy , Roberts Mark J. , Samuelson Larry . 1988 . “Patterns of Firm Entry and Exit in US Manufacturing Industries,” RAND Journal of Economics 495 – 515 . WorldCat Dunne Timothy , Roberts Mark J. , Samuelson Larry 1989 . “The Growth and Failure of Us Manufacturing Plants,” 104 Quarterly Journal of Economics 671 – 98 . Google Scholar Crossref Search ADS WorldCat Egypt. 1907 . Codes Des Tribunaux Mixtes D’Egypte Précédés du Règlement D’organisataion Judiciare . Mourès . WorldCat COPAC Foster Lucia , Haltiwanger John , Krizan C.J. . 2001 . “Aggregate Productivity Growth: Lessons from Microeconomic Evidence.” NBER Studies in Income and Wealth, vol. 63: New Developments in Productivity Analysis, 303 – 63 . Chicago and London: University of Chicago Press. Google Preview WorldCat COPAC Gómez-Galvarriato Aurora , Musacchio Aldo . 2006 . “Larger Menus and Entrepreneurial Appetite: An Empirical Investigation of Organizational Choice in Mexico, 1886–1910.” Paper presented in Session 74 of the XIV International Economic History Conference. Gregg Amanda. 2015 . “Factory Productivity, Firm Organization, and Corporation Reform in the Russian Empire, 1894–1908.” Ph.D. dissertation, Yale University. Grigsby W. E. 1896 . “Mixed Courts of Egypt,” 12 Law Quarterly Review 252 . WorldCat Guinnane T. W. , Harris Ron , Lamoreaux Naomi R. , Rosenthal Jean-Laurent . 2007 . “Putting the Corporation in Its Place,” 8 Enterprise and Society 687 – 729 . Google Scholar Crossref Search ADS WorldCat Guinnane T. W. , Martínez-Rodríguez Susana . 2018 . “Choice of Enterprise Form: Spain, 1886–1936,” 34 Journal of Law, Economics, and Organization 1 – 26 . Google Scholar Crossref Search ADS WorldCat Hansmann H. , Kraakman R. . 2000 . “The Essential Role of Organizational Law,” 110 Yale Law Journal 387 – 440 . Google Scholar Crossref Search ADS WorldCat Herreros Enrique García de. 1914 . Les Tribunaux Mixtes D’Egypte . Alexandria : Molco, Petrini, & Compagno . Google Preview WorldCat COPAC Hilt Eric, , O’Banion Katharine , 2009 . “The Limited Partnership in New York, 1822–1858: Partnerships Without Kinship,” 69 Journal of Economic History 615 – 45 . Google Scholar Crossref Search ADS WorldCat Hoyle Mark S. W. 1991 . Mixed Courts of Egypt . London : Graham and Trotman. Google Preview WorldCat COPAC Jovanovic Boyan. 1982 . “Selection and the Evolution of Industry,” Econometrica 649 – 70 . WorldCat Kuran Timur. 2004 . “The Economic Ascent of the Middle East’s Religious Minorities: The Role of Islamic Legal Pluralism,” 33 Journal of Legal Studies 475 – 515 . Google Scholar Crossref Search ADS WorldCat Kuran Timur. 2011 . The Long Divergence: How Islamic Law Held Back the Middle East . Princeton : Princeton University Press . Google Preview WorldCat COPAC Lamoreaux Naomi R. , Rosenthal Jean-Laurent . 2005 . “Legal Regime and Contractual Flexibility: A Comparison of Business’s Organizational Choices in France and the United States During the Era of Industrialization,” 7 American Law and Economics Review 28 – 61 . Google Scholar Crossref Search ADS WorldCat Lee Yoonsoo , Mukoyama Toshihiko . 2015 . “Entry and Exit of Manufacturing Plants over the Business Cycle,” 77 European Economic Review 20 – 27 . Google Scholar Crossref Search ADS WorldCat Levin Jonathan , Tadelis Steven . 2005 . “Profit Sharing and the Role of Professional Partnerships,” 120 Quarterly Journal of Economics 131 – 71 . WorldCat Levinthal Daniel A. 1991 . “Random Walks and Organizational Mortality,” 36 Administrative Science Quarterly 397–420 . WorldCat Maskin Eric , Tirole Jean . 1988 . “A Theory of Dynamic Oligopoly, II: Price Competition, Kinked Demand Curves, and Edgeworth Cycles,” 56 Econometrica 571 – 99 . WorldCat McAdams David. 2011 . “Performance and Turnover in a Stochastic Partnership,” 3 American Economic Journal: Microeconomics 107 – 42 . Google Scholar Crossref Search ADS WorldCat Michalopoulos Stelios , Papaioannou Elias . 2013 . “Pre-colonial Ethnic Institutions and Contemporary African Development,” 81 Econometrica 113 – 52 . Google Scholar Crossref Search ADS PubMed WorldCat Musacchio Aldo , Gómez-Galvarriato Aurora , Parral Rodrigo . 2008 . “Political Instability and Untimely Dissolution: Partnerships, Corporations, and the Mexican Revolution, 1910–1929.” Working Paper, Harvard Business School. Nicholas Tom. 2015 . “The Organization of Enterprise in Japan,” 75 Journal of Economic History 333 – 63 . Google Scholar Crossref Search ADS WorldCat Olmstead Alan L. , Rhode Paul Rhode . 2016 . “Cotton, Cottonseed, Shorn Wool, and Tobacco—Acreage, Production, Price, and Cotton Stocks: 1790–1999 [Annual], Table da755–765,” in Carter Susan B. , Gartner Scott Sigmund , Haines Michael R. , Olmstead Alan L. , Sutch Richard , Olmstead Alan L. , Sutch Richard , Wright Gavin , eds., Historical Statistics of the United States, Earliest Times to the Present: Millennial Edition , New York, NY : Cambridge University Press . Google Preview WorldCat COPAC Owen Roger. 1993 . The Middle East in the World Economy, 1800–1914 . London : I. B. Tauris . Google Preview WorldCat COPAC Owen Thomas C. 1991 . The Corporation under Russian Law . Cambridge : Cambridge University Press . Google Preview WorldCat COPAC Pamuk Şevket. 2000 . A Monetary History of the Ottoman Empire . Cambridge : Cambridge University Press . Google Preview WorldCat COPAC Saleh Mohamed. 2018 . “On the Road to Heaven: Taxation, Conversions, and the Coptic-Muslim Socioeconomic Gap in Medieval Egypt,” 78 Journal of Economic History 394 – 434 . Google Scholar Crossref Search ADS WorldCat Scott J. H. 1907 . The Law Affecting Foreigners in Egypt: As the Result of the Capitulations, with an Account of their Origin and Development . Edinburgh : W. Green and Sons . Google Preview WorldCat COPAC Stout Lynn A. 2005 . “On the Nature of Corporations,” 2005 Illinois Law Review 253 – 68 . WorldCat Whittle Peter. 1980 . “Multi-armed Bandits and the Gittins Index,” 2 Journal of the Royal Statistical Society: Series B (Methodological) 143 – 49 . WorldCat Whittle Peter. 1982 . Optimization over Time . New York: John Wiley & Sons . Google Preview WorldCat COPAC Wilner Gabriel M. 1975 . “The Mixed Courts of Egypt: A Study on the Use of Natural Law and Equity,” 5 Georgia Journal of International and Comparative Law 407 . WorldCat Wood Leonard G. H. 2011 . “Reception of European Law, Origins of Islamic Legal Revivalism, and Foundations of Transformations in Islamic Legal Thought in Egypt, 1875–1960.” Ph.D. dissertation, Harvard University. Yousef Tarik M. 2002 . “Egypt’s Growth Performance under Economic Liberalism: A Reassessment with New GDP Estimates, 1886–1945,” 48 Review of Income and Wealth 561 – 79 Google Scholar Crossref Search ADS WorldCat © The Author(s) 2019. Published by Oxford University Press on behalf of Yale University. All rights reserved. For permissions, please email: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
TI - Partnership as Experimentation
JF - The Journal of Law, Economics, and Organization
DO - 10.1093/jleo/ewz007
DA - 2019-11-01
UR - https://www.deepdyve.com/lp/oxford-university-press/partnership-as-experimentation-BsLGMXTV5P
SP - 455
VL - 35
IS - 3
DP - DeepDyve
ER -