Characterizing Entrepreneurs

In this section, we hope to articulate how standard microeconomics tends to mischaracterize entrepreneurship by ignoring key features of the decision-making environment. The features we discuss necessitate a different analytical model, which we explore further in the next section.

consider an individual who is contemplating opening a restaurant. What does the decision process look like? While we assume for simplicity that the business is a restaurant, there is no reason to think this deci­sion is unproblematic: many individuals contemplating opening a business consider several types of businesses before settling on one. Moreover, if our potential owner already has a specialized skill, say cooking, there may be a (probably large) differential in opportunity costs between remaining in the food industry versus leaving it. These cost differentials suggest some “path dependence”—i.e., subsequent business or career decisions are influenced by preceding ones—something we discuss in more detail in the section “Inframarginal Choice Plus Time: Path Dependence.” In any case, let us assume our potential entrepreneur has decided to open a restaurant. What next?

An array of additional factors must be decided: exactly what type of restaurant will it be? Will it specialize in ethnic cuisine or feature a “family style” mix of dishes? Will it be formal or casual? If casual, will it feature waiters or counter ordering? The mere decision concerning location is not so simple, as it also bundles together many facets, such as traffic, size and duration of leasing payments, parking, the proximity of competitors, the age and layout of the building, etc.

Approaching the startup process in this way compels us to focus on two features that are ignored in the standard framework: many (if not most) decisions confronting the entrepreneur—as opposed to the manager— involve (i) bundled choices and (ii) computational complexities.

Bundled Choices

We define bundled choices as some meta-decision that necessarily bundles several other subsidiary decisions and we assume that each bundle of decisions constitutes what we have called some “big-picture” state-of- the-world that differs non-marginally from some other decision bundle representing some other big-picture state-of-the-world. A fine Indian restaurant cannot also be a burger shack. The equipment used, require­ments for fixed buildings, and even types of workers in a burger shack are all different from those used in the Indian restaurant. Choosing the type of restaurant is not a little more or a little less of some single, continuously variable input, but all-or-nothing, or either-or, and such a choice requires numerous subdecisions about location, types of kitchen equipment purchased, layout of the restaurant, and recipes used.^[7]

The bundling of decisions is apparent in other contexts. Consider a manufacturer who attempts to establish a new product line. If she completes only a fraction of the bundled tasks—for example, acquiring a new machine but not the skilled technician to run it—the firm will have outlaid resources without any increase in output; in fact, output might fall if regular production was disrupted during the period the new machine was being introduced or integrated into the production process. In this case, the firm might experience negative returns, at least during the period that the entire bundle of tasks (that is, changing from one production process to another) remains incomplete.

Another key implication emerges from both the manufacturing and restaurant examples: entrepreneurs must be unusually committed to their venture and expend considerable effort, at least in part, because entrepreneurial decisions involve a network of actions to be carried out. Leaving work half-done or partially completed (viz. “marginal” adjustments) can lead to worse outcomes than not starting or changing production techniques at all. In other words, once entrepreneurs take the first steps toward developing their businesses, they are likely to confront a series of bundled “all or nothing” choices in terms of their own efforts. If any of these tasks are left unfinished—like our manufacturer or restau­rateur—the venture will remain incomplete. Accomplishing only some of the functions could mean the restaurant never opens its doors at all or the new machine is left gathering dust in the factory's corner. Such an outcome is almost surely worse than never having started down the entrepreneurial path in the first place, which might offer some explanation for the long work hours of entrepreneurs.^[8]

Computational Complexity

We now turn to the second dimension largely ignored in the standard framework: computational complexity. We claim individuals have neither sufficient data nor the cognitive capacity to optimize objective func­tions in the manner assumed by traditional theory. While we could draw on a long line of scholars from different subdisciplines—including, of course, Hayek (1948)—our approach to this issue of data availability and cognitive processing seems to be most compatible with Herbert Simon's “bounded rationality.”^[9] For Simon, bounded rationality is a collection of “theories of how to decide rather than theories of what to decide” (1979, 498, emphasis in original), the latter, of course, being highly descriptive of the mainstream economic approach.

Simon is also deeply skeptical that individuals are able to calcu­late on the margin (a point we develop more rigorously below): “. [the] assumptions [of bounded rationality] about human capabilities are far weaker than those of the [neo]classical theory. Thus,. [bounded rationality makes] modest and realistic demands on the knowledge and computational abilities of the human agents, but [it] also fail[s] to predict that those agents will equate costs and returns at the margin” (1979, 496). Indeed, Simon doubts both that individuals can adequately process the data if they had it, and that they have the data to begin with.¹²

Later, in the same paper, he provides greater insight into his monu­mental critique of neoclassical assumptions and his attempt to replace them with something more realistic:

. The [neo]classical model calls for knowledge of all the alternatives that are open to choice. It calls for complete knowledge of, or ability to compute, the consequences that will follow on each of the alternatives. It calls for certainty in the decision maker’s present and future evaluation of these consequences. It calls for the ability to compare consequences, no matter how diverse and heterogeneous, in terms of some consistent measure of utility. The task, then, was to replace the [neo]classical model

¹²For example, while doing a field study of the administration of public recreational facilities in Milwaukee in the 1930s—which led to the development of his theory—Simon was puzzled as to why the administrative heads

. [did] not, as my economics books suggested, simply balance off the marginal return of the one activity against that of the other.

Our approach is clearly sympathetic to each of these criticisms of the stan­dard model, but we want to emphasize how the problem is exacerbated in an entrepreneurial setting as opposed to a simple utility-maximizing one with a small number of already-determined inputs with readily avail­able prices. Our focus on the bundled, or networked aspect of decisions confronting the entrepreneur suggests there are, quite simply, no easily calculable margins to equate: many decisions that must be made by our imaginary restaurateur—i.e., the type of establishment, location, recipes, input sourcing, decor, etc.—are so complex that there is no “reduced form” equation that would allow for an optimal solution based on equating margins.

A stor y from Amazon illustrates how certain decisions—at the margin—are ignored, sometimes for years, and how easy it is, in retro­spect, to conclude the decision was likely “suboptimal” if taken in isolation from the many other (bundled) decisions of which it was a part. Amazon's long-term effort to eliminate shipping fees is well known, but Wei (2018) documents that, early on, the company had no effective process to determine whether consumers demanding refunds had actually purchased the item from Amazon.

This process underscores an additional challenge that confronts entrepreneurs: they are likely to face nearly constant pressure to make decisions as quickly as possible. The cost of leaving some productive resources idle—as in the case of Amazon or the above entrepreneur intro­ducing a new machine—is often traded off against spending more time analyzing the situation to make a better decision. Decisions must often be made quickly with a globally inferior choice because the quicker deci­sion involves idling fewer resources over time. Making decisions in less time than it requires to fully analyze some problem should be seen as part-and-parcel of the problem of computational complexity.

There are two additional nuances of the computational problem. First, even if our business owner had all the relevant data, she would not have the cognitive capacity to utilize that data optimally. This point implies the decision-making problem is always more complex than simply a “data” problem: whatever data are available must be cognitively processed.^[10] Second, even if she desired to obtain either more data or better cogni­tive processing, our entrepreneur will always confront transaction costs of obtaining more of either; in other words, the decision-maker constantly faces a nontrivial, second-order problem of deciding whether it is worth it to collect more data or develop better skills or education in the pursuit of optimal (or even merely “better”) decisions.

For example, if our restaurateur is trying to decide whether to buy equipment A or equipment B, can she know the optimal level of effort required to research the differences in machines, including searching for reviews, speaking with existing owners of both A and B, and then making third-order calculations about which reviews or owners to weigh more heavily and which to discount? Like Simon (1979), we assert the obvious answer is “no.”^[11] In sum, these three factors—data availability, limits to cognition, and the transaction costs of obtaining more of either—are likely to assure our owner will be unable to truly optimize over the kinds of bundled decisions she faces.

We now return to our earlier point that computational complexity is an important feature of the entrepreneurial setting even if we assume away radical uncertainty. In other words, it is not the uncertainty of how the future will unfold through time that gives rise to the problem of compu­tational complexity in the sense that we describe it here, although the passage of time will most certainly further complicate the complexity problem. It is a problem that exists in a time-invariant (static) environ­ment and arises because no individual can gather all the data that would be necessary to make the kinds of optimal decisions implied by standard theory.

We thus conclude that, in a nontrivial sense, the role of the entrepreneur is overwhelmingly concerned with the choice among a complex set of bundled inputs, such as types of equipment, restau­rant layout, raw materials, and an array of additional factors. it is not concerned with a-little-more-or-a-little-less of a single, divisible input, after the choice of input types has already been made. our approach provides, we would argue, additional microeconomic underpinning for Schumpeter’s notion that it is creative “new combinations” that distin­guish the entrepreneurial function. Further, because the choice between inputs includes many potential options with many interconnected deci­sions, optimizing—that is finding the very best of all possible decisions— as suggested by mainstream theory must be beyond the capacity of entrepreneurs.