1.1 Capital Allocation in Hierarchical Organisations

The purpose of a multi-divisional organisation is to generate more value than any of the individual divisions combined. The whole should be greater than the sum of its parts. Previous work suggests that this is achieved due to factors such as reduced transaction costs (Coase, 1937; Liebeskind, 2000; Teece, 1982, 1980; Williamson, 1981), shared resources (Barney, 1991; Wernerfelt, 1984), increased competitive advantage (Porter, 1980, 1985), increased monitoring (Gertner et al., 1994), and increased synergies (Barney, 1988). The underlying logic is the same: a multi-divisional organisation will be successful if it manages its divisions using processes and resources that are shared or, better yet, are complementary.

In order to successfully manage multiple units, large organisations developed a hierarchical structure. Bower (1970) identified three levels of the typical management hierarchy: business, division, and corporate. These are equivalent to front-line (or bottom), middle, and top level managers (Noda & Bower, 1996). Early theorists suggested that the strategy for the organisation’s growth is driven completely by the top managers; the rest of the organisation simply enacts their proposals. However, Mintzberg and Waters (1985) emphasised the role of an emergent strategy, in which lower level managers affect change in the organisation’s strategy. Other work proposed and found evidence for an iterated process in which a strategic context may be set by top managers, but business projects advanced by lower level managers also contribute to driving the strategy of the organisation (Bower, 1970; Burgelman, 1983; Noda & Bower, 1996).

The way that capital is allocated in an organisation is very important to its growth and longevity. This process is a part of the broader process of resource allocation. A resource can refer to many types of assets that an organisation owns, both tangible and intangible, of which capital is only one (Wernerfelt, 1984). The capital allocation process itself is an important driver of the strategic outcomes of an organisation (Bower, 1970; Bower & Gilbert, 2005), and as a result, is an important influence on its financial performance (e.g., Arrfelt et al., 2015; Bardolet et al., 2010). Sengul et al. (2019, p. 72) describe intra-firm capital allocation as “(i) a process of determination, comparison, and selection among multiple investment alternatives, (ii) taking place across organizational levels of the firm, and (iii) influenced and constrained by the external context in which the firm is situated.” In capital allocation, business-level managers typically formulate project proposals, which their division managers then evaluate. The division managers then choose the projects to send for final approval with the corporate managers. The supply of available capital is also influenced by external sources such as investors, competitors, and customers. However, this thesis focuses on the comparison and selection processes that are relevant during business project evaluation.

Managers ultimately have only limited information about the projects that they evaluate. They typically have access to descriptive information about the investment and its known properties, but also are provided with financial metrics that estimate the returns on the investment. There are many such metrics; they usually attempt to encapsulate a trade-off between predicted future gains, present losses (in the form of the capital spent to pay for the investment), and opportunity costs. Examples include net present value (NPV), internal rate of return, return on investment, cost-benefit, and pay-back period. This thesis focuses on NPV, since it is one of the most frequently used metrics for project evaluation (Graham et al., 2015; Graham & Harvey, 2001; Remer et al., 1993). NPV is the difference between the money that a project is forecasted to make and the initial investment in its development (accounting for the time value of money), as shown in Equation (1.1):

\[\begin{equation} \text{NPV}=\sum_{t=0}^n \frac{R_t}{(1+i)^t}, \tag{1.1} \end{equation}\]

where \(t\) is the time of the cash flow, \(i\) is the discount rate, \(R_t\) is the net cash flow, and \(n\) is the total number of periods. NPV is a useful metric because simply knowing that it is positive suggests that the project that it describes should be profitable. Therefore, metrics such as these have a strong influence on the decision of the manager evaluating a project.

However, there are other influences on project evaluations other than the value of the financial metrics. For instance, politics within or outside the company can lead to situations in which a decision is based on social influence or even manipulation (Garbuio & Lovallo, 2017). Such influence is not necessarily negative; it may involve qualitative feedback from, for instance, a more senior manager (Thamhain, 2014). Research has also shown that the media can have a tangible influence on managerial decision-making (Bednar et al., 2013; Liu & McConnell, 2013). Other sources of influence are the organisational structures and incentives that are in place both externally (Kokkinis, 2019) and internally to the organisation (Rajan et al., 2000; Ullrich & Tuttle, 2004). Such dynamics have also been the subject of economic modelling investigations (Cavagnac, 2005; Ortner et al., 2017; Reichelstein, 1997). Project proposals might also be affected by certain approval structures. For instance, managers might submit overly-optimistic project proposals if they know that the corporate team only accepts projects with a certain minimum NPV forecast.

Another potential organisational influence on capital allocation is the extent of diversification present in an organisation. A diversified organisation is one that possesses different divisions that are unrelated in some way. Penrose (1959/2009, p. 96) defined it as such:

a firm diversifies its productive activities whenever, without entirely abandoning its old lines of product, it embarks upon the production of new products, including intermediate products, which are sufficiently different from the other products it produces to imply some significant difference in the firm’s production or distribution programmes.

Previous work found that organisations that are made up of more related divisions are more successful than those that are made up of unrelated divisions (Harrison et al., 1993; Rumelt, 1974; Shelton, 1988; Wernerfelt & Montgomery, 1988). This is also true within business divisions (Davis et al., 1992). However, more diversified firms have also been shown to be associated with profitability (Grant & Jammine, 1988). This is usually explained by the ability for such firms to avoid risk associated with any one market. Some of the discrepancy in diversification findings has been explained to be due to the specific measures used (Lubatkin & Shrieves, 1986). It may also be because most studies used Standard Industrial Classification (SIC) codes to measure diversification (e.g., Rumelt, 1974), whereas others operationalised it using other approaches (e.g., resource-based; Harrison et al., 1993).

The advantage that related organisations have had has been explained through synergies (Barney, 1988). That is, an organisation with two divisions that can use their resources to better one another are better off together than separately. The 1960s saw a general rise in mergers and acquisitions from executives seeking to diversify their organisations. However, doing so simply for the sake of increasing divisions, rather than an understanding of the possible synergies, leads to the organisation actually being worth less than the sum of its parts (known as a diversification discount; Lang & Stulz, 1994). In fact, many organisations that acquired other businesses to diversify subsequently end up divesting them (Porter, 1987). For instance, in 2018 Australian conglomerate Wesfarmers demerged its Coles division, a successful retailer. Since then, the share price for both companies has risen by approximately 62% and 32%, respectively (Boyd, 2021).

While much of the performance of an organisation depends on influences that are external to the individual managers (e.g., organisational, political), psychological factors are often also quite consequential. For instance, on the one hand, organisational factors such as relevant support teams and approval processes may influence capital allocation depending on the extent of an organisation’s extent of diversification. On the other hand, psychological factors such as ability of managers to compare between business project proposals may also impact allocation differently depending on the organisation’s diversification. It is likely to be more difficult for a manager to evaluate project proposals from two dissimilar divisions that it is to evaluate those from two similar divisions. The organisational influences discussed above often assume that the manager that is making the decisions acts rationally, as per traditional economic theory. However, surveys of executives show that CEOs and CFOs often rely on non-financial factors for capital allocation decisions (Graham et al., 2015). Executives in these surveys identified manager reputation and confidence as two of the most important factors for capital allocation decisions. Further, research in psychology has shown that cognitive biases can influence such capital allocation decisions. Section 1.2 discusses such biases and the relevant implications for the thesis.

1.2 The Psychology of Capital Allocation

Managers of large organisations are generally assumed to have a superior decision-making capability compared with non-managers. However, managerial decision-making involves many of the same processes that have been shown to be affected by psychological biases in the general population (Das & Teng, 1999; McCray et al., 2002; Schwenk, 1984). Further, an organisation’s success ultimately depends on strategic decisions made by top level managers (Mazzolini, 1981). Therefore, despite early work attempting to analyse such decisions using a structured organisational analysis (e.g., Mintzberg et al., 1976), it is important to understand the potential influence of psychological biases on managerial decisions. Research in the field of behavioural strategy has started to do this (Powell et al., 2011).

Psychological research has shown that people tend to make decisions that are inconsistent with neoclassical economic theory. For instance, expected utility theory (EUT; Friedman & Savage, 1948; von Neumann et al., 1944) assumed that people have complete information when making decisions. However, both laypeople and managers of organisations are limited in the amount of information that they have and their ability to use it (Cyert et al., 1956; Simon, 1955). Such inconsistencies with economic prescription are likely to have evolutionary origins, so are sure to be adaptive in certain environments (Bettis, 2017; Gigerenzer, 2008; Haselton et al., 2009). However, there are many situations in which such inconsistency with economic theory can have bad consequences.

Research has shown many ways in which the allocation of capital in an organisation can be influenced by psychological biases. For instance, Benartzi and Thaler (2001) found that people tend to allocate their retirement fund equally between the available options, regardless of their composition. This naive diversification bias was also found in capital allocation for hierarchical firms (Bardolet et al., 2011). Managers allocated capital equally across the available divisions in the firm, regardless of performance. Analysis of real companies found that this behaviour is damaging to firm performance because it means that lower performing business units get subsidised by higher performing units, which are not operating at their full potential (Arrfelt et al., 2015; Bardolet et al., 2010). Subsequent studies found that business unit size also matters; capital allocation to both the smallest and largest units is disproportionate to their actual profitability levels (Bardolet et al., 2017). This was attributed to a combination of naive diversification and political power effects.

Relatedly, people tend to continue expending capital into investments that appear to be failing (Staw, 1981). This escalating commitment is another way that psychological biases can influence allocation in an organisation. This pattern of decision-making is likely a consequence of the sunk cost fallacy, in which people avoid “cutting their losses” even when they know that they cannot recuperate an investment (Parayre, 1995).

Managers also do not always seem to seek profit maximisation. Shapira and Shaver (2014) offered managers and Master of Business Administration (MBA) students two investments from a hypothetical firm: one with the same expected returns as the average of the firm’s current investments and one with lower returns than the firm’s average returns. However, both investments were profitable, so to maximise firm profits both should be chosen. Instead, participants were more likely to only choose the first investment. It seems that the firm’s average returns served as an anchor, so participants did not want to reduce the firm’s average returns, regardless of profitability.

The way that information is presented can also influence allocations. For instance, Yates et al. (1978) showed that people’s evaluations are sensitive to the level of detail in the information provided. They found that people devalued descriptions of university courses more when they had less detail. This may be relevant for managers evaluating project proposals. A proposal might appear more attractive simply due to the level of detail in it, even if the level of detail does not correspond with the quality of each proposal.

Further, people tend to be over-confident in their decisions and forecasts. This has been shown in laypeople (Langer, 1975; Mannes & Moore, 2013; Puri & Robinson, 2007; Soll & Klayman, 2004), as well as in IT professionals (McKenzie et al., 2008) and managers (Barone-Adesi et al., 2013; Kahneman & Lovallo, 1993; Lovallo & Kahneman, 2003). This is important for higher-level managers that evaluate project proposals because the metrics that rely on forecast estimates may be biased by the over-confidence of the lower-level manager that created the proposal. Further, the higher-level manager evaluating the proposal may in turn be over-confident about its prospects due to factors that are unrelated to the underlying value. Overconfidence is also seen when considering the success of projects in hindsight (Bukszar & Connolly, 1988; Christensen-Szalanski & Willham, 1991). This means that it is less likely that managers will be able to effectively learn from both past mistakes and successes due to the potentially erroneous belief that the outcome was anticipated.

Managers often create sensitivity analyses, estimating the worst case, best case, and most likely scenario for a forecast. However, these are likely to be anchored on past experiences that further the manager’s existing beliefs. In fact, prior research has shown that people are poor at constructing subjective probability distributions (e.g., Alpert & Raiffa, 1982; Schaefer & Borcherding, 1973; Tversky & Kahneman, 1974; von Holstein, 1971). Therefore, this suggests that even if the lower-level managers that construct project proposals calibrate their forecasts so that they are not over-confident, they are still likely to provide inaccurate estimates of their degree of confidence.

The above summarises many of the currently known psychological biases related to capital allocation. This thesis focuses on three essential processes within the capital allocation process: (a) risky choice, (b) the comparison between diversified businesses, and (c) the influence of prior experience. Each of these is prone to separate biases, that are also interrelated. The subsequent subsections review the literature for these processes.

1.2.1 Risky Choice

Neoclassical theories such as EUT suggest that when faced with multiple risky options people should choose the option with the highest expected value (EV), all else being equal. This means multiplying the value of each option by its probability and comparing the resulting values (first documented in Pascal, 1670/1999). For instance, imagine being presented with the following two choices:

1. a gamble that involves a 50% chance gaining $200 and a 50% chance of losing $100; or

2. gaining/losing nothing.

In option A, the EV is calculated as \(200 \cdot 0.5 - 100 \cdot 0.5 = 50\). Since the EV for option A (50) is higher than the EV for option B (0), EUT would suggest that option A should be chosen.

This basic principle was extended by Bernoulli (1738/1954), who suggested that a person’s subjective value of money differs depending on their current wealth. This diminishing marginal utility suggests that the more money a person already has, the less value acquiring more money will have for him. For example, the experience of a rich man that finds $10 on the street is very different to the experience of a homeless man that finds $10 (Bradley, 2013). Even though $10 was gained in both cases, $10 has less value to a person that already has, for example, $1,000, than for a person that initially only has $10. This principle is usually modelled as an power function (with a fractional exponent).

Prospect theory (Kahneman & Tversky, 1979; Tversky & Kahneman, 1992) challenged EUT by suggesting that people’s subjective value of money does not depend on their state of wealth—it depends on a change of wealth from a reference point. This is important because people’s subjective value of money is different depending if they are gaining or losing money. Specifically, losses have a stronger psychological impact than equivalent gains. This disparity is one of the most settled and consistent findings in psychology and economics, having been well-replicated (e.g, Ruggeri et al., 2020). The fact that losses loom more than equivalent gains for the vast majority of people is referred to as loss aversion (Kahneman & Tversky, 1979). This finding was the primary reason that Daniel Kahneman won the Nobel Prize in Economics in 2002 (Kahneman, 2003). Loss aversion has been found with small amounts of money in experimental settings (Kahneman & Tversky, 1979; Tversky & Kahneman, 1992) and with millions of dollars in corporate settings (Koller et al., 2012; Swalm, 1966). The effect has been found in young children (Harbaugh et al., 2001), the numerous disparate cultures in which it has been tested (Weber & Hsee, 1998), and even in capuchin monkeys (Chen et al., 2006). Furthermore, a neural basis for loss aversion was identified (Tom et al., 2007). Therefore, loss aversion is clearly central to human cognition and behaviour.

The function that represents the value of a prospect describes both loss aversion and diminishing marginal utility, as shown in Equation (1.2):

\[\begin{equation} v(x) = \left\{ \begin{array}{l} x^\alpha \text{ if } x \ge 0 \\ -\lambda(-x)^\beta \text{ if } x < 0, \tag{1.2} \end{array} \right. \end{equation}\]

where \(x\) is the possible outcome, \(\lambda\) represents the loss aversion coefficient, and \(\alpha\) and \(\beta\) represent the diminishing marginal utility for gains and losses, respectively.

In other words, loss aversion means that losses have more impact than equivalent gains. In fact, the impact of loss aversion can be expressed even more precisely, as a measurement of the ratio of the slopes of the curve for gains and losses. This measure tells us the average amount that losses have more impact than equivalent gains. In a sequel to the original prospect theory paper, Tversky and Kahneman (1992) measured a median coefficient (\(\lambda\)) of 2.25 of loss aversion. This means that people respond to losses 2.25 times more than equivalent gains. Similarly, this paper measured a median exponent (representing diminishing marginal utility, \(\alpha\) and \(\beta\)) of 0.88 for both gains and losses. This means that people discount money the more of it they have by a rate of \(x^{0.88}\).

Figure 1.1 shows loss aversion as the function being steeper in the domain of losses than the domain of gains. It shows diminishing marginal utility by the slight curve of the function. Equivalent changes in actual wealth from the references point (x-axis) have different impacts on the changes’ subjective value (y-axis). An increase in wealth (\(x = 1\)) brings about an equivalent increase of value (\(y = 1^{0.88} = 1\)). However, a decrease in the same amount of wealth (\(x = -1\)) brings about a decrease in value 2.25 times the value of the equivalent gain (\(y = -2.25 \cdot (-(-1))^{0.88} = -2.25\)).

Figure 1.1: An example of the value function in prospect theory.

This research is relevant to capital allocation because the project proposals that managers evaluate invariably involve an element of risk. Therefore, managers are likely to be affected by similar effects on risk that have been shown in laypeople. However, hierarchical organisations offer an even more complex situation. Lovallo et al. (2020) found that the risk profiles of lower-level managers are lower than those of the top managers. They suggest that this may be due to lower-level managers’ loss aversion to accepting projects that may jeopardise their job. However, the top managers recognise that a loss in one or more business units is likely to be offset by gains in other units. Such an inconsistency in risk profiles across the levels of an hierarchical organisation fails to take advantage of the benefits of risk aggregation, which has long been understood in external markets (Markowitz, 1952).

Lovallo et al. (2020) suggested that lower-level managers’ failure to aggregate risk to the degree desired by top executives is costing companies approximately a third of the total EV of new project proposals. This is an example of a negative consequence associated with ignoring statistical concepts such as risk aggregation. It is thus critical to identify ways to support risk aggregation across organisational hierarchies. The psychological literature shows that people’s risk aggregation is facilitated through various choice bracketing manipulations. However, there has been no work that investigated such situations without providing participants with feedback in between decisions; this critically limits the external validity of this work because in the real world, organisations evaluate several projects before seeing the outcomes of any one decision. The experiments presented in Chapter 2 investigate the effects of choice bracketing on risk aggregation without feedback.

1.3 Chapter Overview

In sum, the potential consequences of a diversified hierarchical structure are that business projects will be considered one at a time, and if they are considered together, disparate project types will make comparisons hard. Considering projects one by one might mean that risk is not aggregated across projects and therefore value is lost. The difficulty to compare will lead to both potentially relying on unreliable metrics, and relying on improper anecdotal evidence. The thesis is that people often go half-way. They do not completely disregard the normative strategy, but also struggle to use statistical concepts such as aggregation, variance, and sampling.

The previous section identified three capital allocation processes that are currently under-studied and so are important to investigate further. First, the evaluation of individual project proposals may lead to managers only considering such projects one at a time, despite the opportunity of aggregating a portfolio of such projects. The choice bracketing literature suggests that there are ways of facilitating such aggregation, but does not investigate this without providing participants inter-trial feedback. Second, in situations in which managers compare multiple projects, the structural alignment literature suggests that managers in diversified firms will struggle to allocate capital, more than those in more integrated firms. Further, these managers may not be sensitive to the variance inherent in the financial metrics they rely on. Third, a difficulty to compare across existing projects may instead mean a reliance on prior case studies from personal or external experience. Research on anecdotal bias suggests that managers may rely more on such case studies than on aggregated data, but it is unclear whether their decisions will depend on anecdote relevance. Further, it is unclear if they will appropriately use information about the anecdote’s sample distribution.

The rest of this thesis investigates the psychology of capital allocation decisions in three chapters that describe empirical work, two theoretical chapters, and a general discussion chapter. Chapter 2 describes two experiments that investigate the effects of choice bracketing on risk aggregation without feedback. Chapter 3 is a short theoretical chapter that discusses the difference between evaluating project proposals with inherent budget estimates and the process of allocating an existing budget top-down. Chapter 4 describes three experiments that investigate the effects of alignment and reliability type—verbal or numerical—on allocations. Chapter 5 is another short theoretical chapter that discusses the trade-offs that people make when using information to evaluate project proposal options. Chapter 6 describes two experiments that investigate the effects of anecdote similarity on the anecdotal bias. Finally, Chapter 4.2.3 discusses the theoretical and practical implications of the empirical chapters and concludes the thesis.