Toward a political economy of complex interdependence

Introduction

How do we theorize about the global political economy in an age of complex interdependence? Historically, international political-economy scholarship has conceptualized interdependence as increasing connectivity with costly consequences between national economies generated by rising cross-border flows of goods, services, money, and people (Keohane and Nye, 2001: 7–9). Yet, although connectivity with costly consequences constitutes one important element of interdependence, connectivity alone fails to adequately characterize interdependence. The connectivity also has structure. Consider global financial interdependence. Contemporary global financial interdependence has a strongly hierarchical structure (Kubelec and Sá, 2012; Oatley et al., 2013). The system is one in which a very small number of national economies occupy roles as global financial centers, attracting capital from and intermediating capital flows between counterparties across the entire world. In contrast, most other countries are connected to the center but are only loosely connected directly to one another. Moreover, this hierarchical structure is persistent; the global distribution of the stock of cross-national portfolio assets and liabilities is stable from one period to the next and is further reinforced by additional flows. Finally, the persistent hierarchical structure exhibits rising heterogeneity across a variety of scales. Some financial institutions have developed into important global banks, while most others focus most heavily on national or even subnational business. Some countries have emerged as global or regional financial centers, while most have not. Rising differentiation between financial institutions and national financial systems has led to greater heterogeneity in banking regulation; current rules, for instance, treat systemically important financial institutions (SIFIs) differently than they treat non-SIFIs. Complex interdependence in the global financial system is thus characterized by a persistent structure of connectivity and rising heterogeneity.

Complex interdependence is reshaping the political economy of finance. Increasingly, developments within societies are tied to the way in which local financial institutions are embedded in and seek to exploit their connections to the global financial system. Icelandic banks, for example, could never have become so highly leveraged between 2006 and 2008 had they not been able to tap into international markets. More broadly, whether an emerging market economy experiences a credit boom, asset bubble, and banking crisis often has more to do with developments in the center of the global financial system than with regulatory characteristics or other dimensions of domestic politics (Bauerle Danzman et al., 2017). Moreover, the central principles that underpin domestic banking regulation in most national economies are established through an international process centered upon the Bank for International Settlements (BIS) and dominated by actors who represent the interests of the major global financial centers, the US especially and the European Union (EU) occasionally (Newman and Posner, 2018; Posner, 2009). Consequently, private actors have mobilized around the BIS in order to shape these international regulatory frameworks. Therefore, complex interdependence has transformed the politics of global finance such that one can no longer readily identify distinct domestic and international levels of analysis that one might study in isolation from one another.

Understanding contemporary complex interdependence, both in the specific context of the global financial system and more broadly in International Political Economics (IPE), requires us to develop new theory. Existing IPE theory, based largely on the logic of the Open Economy Politics (OEP) framework, helps us understand what agents want and how they interact within the parameters of a particular system. However, these mainstream theoretical models are not very useful for understanding contemporary complex interdependence because they do not offer analytical purchase on the behavior of systems or into questions about change. ¹ That is, if we are to understand global financial interdependence, we must develop theories about systems that extend well beyond the state-centered framework that dominates contemporary American IPE and International Relations (IR). Moreover, the international financial system has changed over time in fundamentally important ways. It does us no good to assume that the contemporary global financial system operates in the same way today as it did in 1950, and therefore that theories developed during the 1960s help us understand global finance in 2020. In order to theorize about the performance and evolution of the global financial system, we must embrace new (to us) theoretical logics.

I argue that in order to enhance our understanding of global financial interdependence, we need to draw heavily from the complexity sciences as motivated by evolutionary logics and rely less than we do now on theoretical metaphors drawn from Newtonian mechanics. The complexity sciences encourage us to recognize the plasticity of social systems — the attributes of the actors that populate a system, the characteristics of the interactions between these actors, the structures that these interactions create, and the outcomes generated by the interaction among these components are all persistent rather than permanent features. Thus, central characteristics of the system — the type of actors that inhabit the system, how the system is structured, and how the system works — all evolve over time as a consequence of the actions and interactions of the actors that inhabit it.

This article draws on these theoretical logics to articulate a Political Economy of Complex Interdependence (PECI). I proceed in four steps. The first step introduces complexity theory, highlights three key assumptions that it makes about social systems, and contrasts them to the equivalent assumptions embodied in the mainstream OEP approach. The second step highlights what these alternative assumptions imply for how we conceptualize and study global political economy. The third step illustrates a PECI approach by focusing on the development of the subprime crisis. Finally, I consider the epistemological implications of PECI and offer concluding comments.

Core elements of a political economy of complex interdependence

A fairly large existing literature encourages IR scholars to draw more heavily from complexity science (Albert et al., 2010; Axelrod, 1997; Bousquet and Curtis, 2011; Cederman, 2003; Gunitsky, 2013; Harrison, 2006; Hoffmann and Riley, 2002; Jervis, 1997; Kavalski, 2007; Ma, 2007). Enthusiasm for complexity science reflects a perception that the core assumptions that this approach makes about social systems fit better with the world we study than the Newtonian ontology that informs IPE’s mainstream perspectives. Although complexity science is large and varied, most who would consider themselves to be engaged in complexity science would agree that the field rests on three core propositions. First, adherents argue that “more is different” (Anderson, 1972). Higher-level entities are not merely the sum of the characteristics of the lower-level entities of which they are constituted. For instance, a financial system is different than the typical balance sheet of the financial firms that operate within it. More is different because aggregating induces interaction among actors, and interaction creates a larger possibility space — the set of events that may occur is thus expanded as a consequence of interaction (Clayton, 2013: 340). Second, complexity scientists believe that social systems change as agents, information, and the environment coevolve. Agents develop specific characteristics within the context of a specific environment. Agents learn about the environment, and as they use this knowledge in pursuit of their objectives, they often change the environment. In turn, the new environment reshapes agents. Third, social systems are indeterminate and unpredictable. As social systems contain a large possibility space, and because this possibility space changes over time, humans have limited capacity to predict outcomes. As Boulding (1987: 116) notes, this inability “to predict is not the deficiency in human knowledge, but an inherent and inescapable property of the system itself.”

Although complexity theory seems especially well-suited as a framework for IPE, the approach remains firmly on the margins. ² The inability of complexity theory to establish a beachhead in IPE is surprising because many of the most prominent applications of complexity theory focus on economics and financial markets (Arthur, 2015; Mandelbrot, 2006; Sornette, 2003). My purpose is not to apply complexity mechanisms uncritically to IPE, but to articulate an analytical framework for IPE that is motivated by three core characteristics of complexity sciences and to highlight three practical implications that this framework has for how we study IPE. Table 1 helps orient the discussion. I articulate the PECI framework and compare it to the OEP perspective along three dimensions: the type of system; assumptions about human rationality; and system dynamics. I then articulate three practical implications of PECI and contrast them to OEP-based research: the unit of analysis; the importance of distant causes; and the study of change.

OPEN IN VIEWER

Table 1.

Assumptions and characteristics of alternative perspectives on political economy.

	PECI	OEP
Motivating scientific metaphor	Evolutionary biology: this reflects a broad recognition of open systems that change over time in unpredictable ways. Understanding of the present system confers no predictive power about future states of the system.	Newtonian mechanics: closed systems that operate mechanically, even if probabilistically rather than deterministically. Once we understand the laws that govern behavior, we can predict how the system will work in the future.
Character of system	Complex system. Not decomposable.	Simple mechanical system. Modular and decomposable.
Human behavior	Inductive rationality in environment of pervasive uncertainty.	Rational maximizers in environment of known risk.
Ergodicity	The international political economy is an open, non-ergodic system.	The international political economy is an ergodic system.
Unit of analysis	System; relations or network structures. Theorize system dynamics: dynamics on a given network/system; dynamics that drive change of a given system.	Individual, though the unit varies (state, firm, person).
Causal relationships	Global relationships are increasingly important.	Local relationships are most important.
Studying change	Emphasizes endogenous sources of rising systemic complexity.	Models tend to emphasize equilibrium; change is typically exogenous.

PECI: Political Economy of Complex Interdependence; OEP: Open Economy Politics.

First, PECI conceptualizes the global political economy as a complex system. A complex system is “a system in which large networks of components with no central control and simple rules of operation give rise to complex collective behavior, sophisticated information processing, and adaptation via learning or evolution” (Mitchell, 2009: 13). Both components of the term “complex system” require elaboration because both have specific meanings that differ from how they are commonly used in IPE. System does not mean “system level” as defined by the levels-of-analysis approach of standard IR theory. In standard theory, system-level theories focus on the interaction between states, and “black-box” domestic politics and individual characteristics. This is not what PECI means by system. In very broad terms, PECI defines a system as “any structure that exhibits order and pattern” (Boulding, 1985: 9; see also Simon, 1962). Thus, to say we should study the global political economy as a system is to say that we should study the global political economy as a coherent structure that exhibits order and pattern.

More broadly, PECI characterizes a system as a pattern of relationships between private and public agents that extends within national economies as well as across national borders and the associated institutions, regulations, and organizations within this environment. It is often useful (but is not necessary) to conceptualize complex systems as networks (Bousquet and Curtis, 2011: 46). ³ The global financial system, for instance, is a network of relationships between large institutions, some private some public, individuals, and public authorities residing in multiple jurisdictions. Small banks in rural America, for instance, are tied (directly and indirectly) to the People’s Bank of China, which is, in turn, tied directly and indirectly to (among others) the Federal Reserve, the US government, and the European Central Bank.

“Complex” also has specific and consequential significance. Often, IPE uses the word “complex” as a synonym for complicated, or as an umbrella term that covers a variety of departures from the standard linear and additive model of causality (Braumoeller, 2003; Chaudoin et al., 2015). Without questioning the utility of that usage, complex in this context connotes something other than “complicated” and something broader than non-linear and non-additive causation. One might characterize complexity in this sense as the ability of a system to generate surprising events — outcomes that were not expected by even a well-informed observer of the system (Mitchell, 2009: 54). Systems can be remarkably complicated without being the least bit complex. Mechanical systems, for instance, have many moving parts and are therefore quite complicated (the Airbus A380 comes to mind). However, in their normal operating mode, they do not generate surprises. Therefore, such mechanical systems are not complex systems. Financial markets, in contrast, exhibit considerable complexity. The ability of agents to accurately predict the future value of an asset is limited and asset prices move in directions and by amounts that are often very unexpected. Similar dynamics are evident in the operation of the macroeconomy (Blyth and Matthijs, 2017). A similarly high degree of uncertainty surrounds the broader global financial system; the 2008 crisis, for instance, was unexpected (“a 20 sigma event” is how one investment banker is alleged to have characterized it) (Jones, 2008). To say that the global political economy is a complex system, therefore, is to assert that it is a structure of relationships that stretches within and spans across societies, and whose dynamic characteristics are such as to generate unexpected outcomes.

This is a very different conception of a system than that which underpins OEP. OEP conceptualizes a system in Newtonian terms as a closed mechanical system. ⁴ A mechanical system is one that is constructed from a potentially large number of sub-components but that can be understood by examining these sub-components in isolation from each other. An analog (mechanical) watch offers a prototypical example of a simple mechanical system. Such a watch is constructed from five sub-components, and the function that each sub-component performs is in no way determined by its relationships with others. The mainspring, for instance, has neither more nor less potential energy when incorporated into a watch than it has when it rests on the watchmaker’s workbench. Although I doubt few scholars believe the international system is actually a watch-like mechanism, OEP scholarship relies upon the logic of simple mechanical systems. Lake (2011: 473), for instance, writes that, “ideally, we want modular theories — separate, self-contained, and partial theories — that connect more or less well to other theories to carry out larger explanatory tasks,” and “the broadly shared assumptions allow the components to be connected together into a more complete whole” (Lake, 2009: 225). Chaudoin, Milner, and Pang (2015: 304–305) seemingly concur when they suggest that:

employing the most complex model of interaction is not necessarily the best path to take, especially as an initial step. It may be much more informative to start simply and then see if the addition of complex interactions changes the explanation substantially.

Second, PECI assumes that humans are purposive actors but that they often cannot pursue optimal strategies. Two factors combine to make it difficult for agents to optimize (Bednar and Page, 2007, 2016; Bednar et al., 2012). The first factor is the complexity of the system itself. Uncertainty about the strategies that other agents are likely to choose increases in line with the complexity of the interaction. As Arthur (2015: 4) notes:

if I cannot know exactly what the situation is, I can take it that other agents cannot know either. Not only will I have to form subjective beliefs, but I will have to form subjective beliefs about [the] subjective beliefs [of other agents]. And other agents must do the same.

In other words, the cognitive load associated with selecting the optimal course rises in line with complexity of the system. The second factor is cognitive limitations: as complexity rises, agents become increasingly unable to locate and implement optimal action (Bednar et al., 2012: 14). ⁵ Instead, in the context of rising complexity, actors employ heuristics and societies develop culture as aids to their decision-making. ⁶

A growing literature has established that cognitive limitations have led to the development and use of individual heuristics and community-wide cultures (Bednar and Page, 2007, 2016; Cosmides and Tooby, 2013; Samuelson, 2001). Such heuristics can be individual devices or they can be shared. In both cases, they arise as a consequence of learning: agents posit multiple hypotheses and test them as they engage with the world and with other agents. They embrace those that succeed and reject those that do not (Arthur, 1994). We might think of proprietary trading strategies that individuals develop and refine through trading on their own (or their employer’s) account as an example; the Black-Scholes model most famously, but also the Gaussian Copula Model (GCM) at the center of the subprime crisis and the whole set of trading algorithms more broadly, exemplify such heuristics. Successful heuristics are written down and transmitted across societies and (sometimes) across generations. We can characterize these as culture: a common behavioral rule that “guides strategies of action” — persistent ways of ordering action through time (Bednar and Page, 2016: 454). We might treat Keynesian economics, for instance, as one such heuristic device that emerged within a particular context as an attempt to understand how to manipulate the world to produce better economic outcomes.

PECI’s focus on adaptive behavior differs from typical OEP assumptions about individual decision-making. First, OEP rarely problematizes the structure within which interaction occurs, treating it as (or merely assuming that it is) an objective reality. Consequently, agents are not uncertain about the world or about the consequences of their actions. Second, all possible outcomes from a choice are known and agents can assign probabilities to every outcome. ⁷ Third, and because of the first and second assumptions, agents maximize. As one recent survey of the standard rational choice model summarized: “Actors correctly understand their environment and update their beliefs appropriately, given the information available to them. And given the environment, their preferences, and their beliefs, actors make choices that will, probabilistically, return the greatest utility” (Hafner-Burton et al., 2017: S7). Mainstream OEP research typically assumes that observed departures from the idealized rational actor model occur as a consequence of individualized psychological characteristics (“dispositions rooted in emotion, social psychology, and even genetic differences” (Hafner-Burton et al., 2017: S2)) rather than because of the complexity generated by the dynamic interdependencies of a complex system.

The third aspect of PECI that is shaped by complexity science is the assumption that the global political economy is non-ergodic: it changes as it moves through time. ⁸ The assumption of non-ergodicity is central to neo-institutionalism. As Douglass North (1999: 2) has written: “the world with which we are concerned is continually changing, is continually novel…. [F]or an enormous number of issues that are important to us, the world is one of novelty and change; it does not repeat itself.” ⁹ Paul David offers a more precise conceptualization. Ergodic processes are ones “whose dynamics guarantee convergence to a unique, globally stable equilibrium configuration; or, in the case of stochastic systems, those for which there exists an invariant (stationary) asymptotic probability distribution that is continuous over the entire feasible space of outcomes” (David, 2001: 4–5). In a non-ergodic system, in contrast, there is no “globally stable equilibrium configuration” toward which the process converges. A non-ergodic stochastic process “is one whose asymptotic distribution evolves as a consequence (function of) the process’s own history” (David, 2001: 5). Thus, the distribution of outcomes in the global political economy changes over time as a consequence of the adaptive interaction among the agents who jointly constitute the system. As a result, the global political economy of today differs from the global political economy of yesterday.

In the global financial system, novelty emerges from adaptive behavior. The general principle is widely recognized, even if not always framed expressly in these terms. For example, Goodhart’s Law (and the Lucas Critique similarly) highlights the instability of cause–effect relationships in human systems. Charles Goodhart’s initial formulation was prompted by the Bank of England’s frustration with its effort to control the money supply. The law asserted that “Any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes” (Goodhart, 1994: 96). Strathern (1997: 308) offered a more general formulation: “when a measure becomes a target it ceases to be a good measure.” She applied the law to audit systems in British higher education, arguing that once authorities select some metric to measure academic performance, the metric ceases to reliably discriminate between individual performances. Others have applied the law to financial systems: “the setting of any particular rule will invite regulatory arbitrage or encourage innovation to circumvent rules” (Sheng and Looi, 2003: 237). Each of these examples illustrates a non-ergodic process. An effort to manage or manipulate a social system using an approach based on inferences drawn from a distribution of observed (and thus historical) behavior causes individuals to change their behavior, thereby ensuring that the prior distribution ceases to be representative of human behavior. The probability distribution changes as a function of the process’s own history.

OEP, in contrast, assumes that the global political economy is ergodic. OEP inherits its ergodicity assumption from its attachment to neoclassical economics. As Paul Samuelson (1968: 12) wrote 50 years ago, economists embraced ergodicity because:

we theorists hoped not to introduce hysteresis phenomenon into our models as the Bible does when it says “We pass this way only once” and, in so saying, takes the subject out of the realm of science into the realm of genuine history.

OEP embraces this logic. Lake (2011: 474) characterizes mid-range theory in expressly ergodic terms. The approach “builds generalizable explanations from covering laws … as statements of enduring relationships.”

As a framework for analysis, therefore, PECI differs from OEP in three important ways. First, PECI assumes that the global political economy is a complex system rather than a mechanical one. Second, PECI assumes that humans create and sustain this complex global political economy but cannot predict how the system will behave over time. This uncertainty is due, in part, to the magnitude of the possibility space amplified by the intersubjective process through which this possibility space is generated. Finally, PECI assumes that the global political economy changes as it moves through time as a consequence of human interaction.

Implications of the core assumptions

PECI has implications for how we theorize about (financial) interdependence that differ sharply from the theories that we more commonly apply. I focus here on three of the most important implications.

First, for PECI, the system is the unit of analysis. In broad terms, this means that phenomena of interest are caused by the structure of relationships among agents and by the interaction between agents within this structure rather than by the essential characteristics of particular units. Moreover, to the extent that particular unit attributes such as institutions and regulatory regimes influence outcomes of interest, these characteristics matter only through their interaction with agents. PECI thus encourages us to study how the global political economy is organized, how this organization shapes system behavior and performance, and how the system changes over time. For example, PECI encourages us to study the network topography of the global political economy. Here, the degree of system hierarchy may condition the dynamics of contagion and diffusion (Oatley et al., 2013). Flat systems may be more vulnerable to contagion than hierarchical systems. PECI encourages exploring whether and how a state’s position within the system conditions the local impact of foreign shocks and causes otherwise identical shocks to produce different global repercussions. For example, recent work indicates that systemic banking crises that occur in the center of the global financial system (the US) have global consequences, while banking crises that occur in the EU and emerging markets have predominantly local consequences (Bauerle Danzman et al., 2017). Finally, studying the system implies theorizing about how and why the system’s structure and performance change as it moves through time.

PECI’s focus on the system as the unit of analysis differs from standard practice in OEP. OEP has dedicated relatively little attention to studying the international system over the last 20 years (Cohen, 2008, 2016). OEP has devoted more attention to the domestic determinants of policy choice and to the origins of individuals’ economic policy preferences. To be clear, this focus on domestic and individual-level characteristics is not intrinsic to the OEP framework, but instead reflects a preference for such work. Nevertheless, OEP research has focused on individuals and states because for OEP, the individual rather than the system is the appropriate unit of analysis. As Lake (2009: 225) says: “OEP begins with individuals, sectors or factors of production as the unit of analysis.” One must first establish empirically accurate micro-foundations and build upon them to develop explanations of higher-level outcomes. Furthermore, even when states rather than individuals become the unit of analysis, scholars typically isolate states from the international system and treat them as if they were independent entities. ¹⁰ Thus, where PECI encourages us to study the system and agent as mutually constitutive and thus inseparable, OEP holds that we cannot understand the system until we understand the origins and content of individual preferences. ¹¹

Second, because the system is the unit of analysis, PECI encourages us to recognize that causes of the local outcomes of greatest interest are often global. PECI encourages us to look for the causes within the dynamic characteristics of the system. For example, the local manifestation of a systemic banking crisis can result from events that occur in other regions of the world. Contagion is the most well-known mechanism, wherein a systemic banking crisis that originates in Cyprus can spread to other European banking systems as a consequence of cross-border inter-bank exposures. Diffusion is another mechanism through which distant causes bring about local effects (Braun and Gilardi, 2006; Elkins et al., 2006; Gilardi, 2012; Simmons and Elkins, 2004; Simmons et al., 2006). Local banking regulations in most high- and middle-income countries, for instance, have been shaped by the diffusion of standards from the center (the US and EU) to the Basel Committee on Banking Standards and then from the Basel Committee outward. Although contagion and diffusion are familiar systemic mechanisms, they hardly exhaust the ways in which global causes have local consequences. A growing body of research finds that the probability of a banking crisis in the emerging market economies (EMEs) varies as a function of the magnitude of the US macroeconomic imbalance (Bauerle Danzman et al., 2017; Forbes and Warnock, 2012; Schwartz, 2009). During periods in which the US records a large macroeconomic imbalance, and thus borrows heavily from the rest of the world, the probability of banking crises in EMEs falls quite sharply. Moreover, as the US moves back toward macroeconomic balance, the probability of crises in the periphery rises.

OEP, in contrast, firmly embraces Tobler’s First Law of Geography: “everything is related to everything else but near things are more related than distant things” (Tobler, 1970: 234). The OEP belief that the most important causes of phenomena of interest are most often found in close proximity is apparent in two characteristics of OEP theories. First, the overwhelming majority of OEP theory elaborates causal relationships between attributes at the same level of analysis. Whether a government fixes or floats its national currency depends upon domestic political and institutional characteristics. The ability of a state to attract foreign direct investment is a function of the domestic rule of law and the presence of a bilateral investment treaty that ties the hands of the state. Second, OEP further localizes cause and effect relationships by searching for the cause of specific outcomes inside the same issue area. Banking system crises are caused by the particular characteristics of domestic bank regulation. Exchange rate policy is caused by firms’ preferences over exchange rate stability and currency values.

Third, PECI encourages theorizing about systemic change. In abstract terms, PECI is primarily interested in theorizing about variation in systemic structure and behavior as the global political economy moves through time. Blythe and Matthijs (2017) offer an excellent recent example. They characterize the global macroeconomy as “an evolutionary system driven by dynamics of inflation and deflation” (Blythe and Matthijs, 2017: 205) and explore how through their “normal operation,” the institutional pathologies of macroeconomic regimes gradually erode the regime’s effectiveness, thereby creating support for a shift to a new regime. As they explain:

as institutions became more tightly coupled to one another, with feedback loops from one set of institutions impacting others in unexpected ways, any “normal” policy intervention began to demand further second-order correcting interventions to steer the system, that then in turn created further feedback loops and increased the demand for more interventions. (Blyth and Matthijs, 2017: 214)

The New Interdependence Approach (NIA), which endogenizes the evolution of global rules as a function of the interaction between globalization and domestic politics, constitutes another promising line of research (Farrell and Newman, 2014, 2016). More broadly, we need theories that endogenize systemic change by examining how agents adapt their behavior to the (often unanticipated) consequences of decisions taken and then institutionalized in prior periods, and how these adaptations push the system over time, either continuously or in large discrete jumps.

PECI’s focus on systemic change constitutes a major departure from OEP research. OEP’s assumption that the system is ergodic creates little reason to search for theories that endogenize systemic change. The goal of inquiry is to identify causal relationships that hold across time and space. OEP is obviously not blind to changes that occur in the global economy, but the framework assumes that major changes in the characteristics of the system are brought about by societal and governmental reactions to exogenous shocks. Farrell and Newman (2016: 714) note that “Many practitioners of … OEP … see globalization primarily as an exogenous shock that activates domestic interest groups and in turn is filtered through domestic institutions to either support or oppose economic openness” rather than as an outcome of agent interaction that has consequences to which agents adapt their behavior. Similarly, many practitioners of OEP see financial crises as exogenous shocks that activate domestic interest groups (see, e.g., Mosley and Singer, 2009) rather than as outcomes that emerge endogenously from agent interaction over regulatory regimes, macroeconomic policy regimes, and global imbalances (Oatley, 2015; Schwartz, 2009).

PECI and the subprime crisis

Understanding global financial interdependence requires us to study system processes and systemic change, and in order to make progress along this path, we could usefully rely more heavily on PECI-informed theory and rely correspondingly less on OEP. To illustrate this claim, I trace the coevolution of financial innovation and financial regulation that was central to the development of the subprime crisis of 2008–2009 and show how this process worked to concentrate risk at the center of the system rather than distribute risk as intended. I focus on the subprime crisis because it constitutes the single most important event generated by contemporary global financial interdependence and because, more broadly, financial crises are to political economy what wars are to international relations: both constitute a looming threat of systemic malfunction that almost never occur, and when one does occur, it happens in a way that policymakers largely failed to anticipate. Consequently, even research that does not focus on crises per se should employ a theoretical perspective that can accommodate the processes through which they emerge.

The subprime crisis is, in many respects, a story about the emergence, widespread embrace, and collapse in value of Collateralized Debt Obligations (CDOs) constructed from Mortgage-Backed Securities (MBSs). This trajectory resulted from a set of coevolutionary processes that changed the financial system and from the inability of market participants and regulators to acquire information about the nature and extent of these changes. A CDO is an instrument that allows institutions to manage risk through pooling and tranching (Coval et al., 2009). Pooling assembles a bundle of assets, such as subprime mortgages, each of which individually carries a relatively high credit risk but which are unlikely to default simultaneously at a high rate. Tranching creates a hierarchy of claims on the income generated by this underlying pool. A typical CDO is divided into three tranches: senior, mezzanine, and equity. The equity tranche absorbs initial losses from defaults. Once the equity tranche is eliminated, the mezzanine tranche begins to absorb losses. The senior tranche is jeopardized only once the mezzanine tranche has been exhausted. In a properly designed CDO, therefore, the senior tranche is very well protected by the subordinate tranches, and this protection justified the AA and AAA ratings assigned by the credit rating agencies (CRAs).

CDOs constructed from subprime mortgages did not exist prior to 2000, and they came to play a central role in the crisis as a result of the coevolution of financial innovation and financial regulation. CDOs emerged as part of the broader revolution in risk management that took hold in finance during the 1980s. Traditionally, an originator of a mortgage (or other asset) carried the credit risk on its balance sheet until maturity. After 1990, financial institutions developed models and instruments that allowed them to move this risk off their balance sheets by dividing it up and selling it on to investors willing to hold it. CDOs and the underlying models used to construct them developed in the context of and became an important element of this risk revolution. JP Morgan developed an early version of the GCM in the mid-1990s (MacKenzie and Spears, 2014a: 404). In 1999 and 2000, David X. Li drew on actuarial research, where copula functions are used to model the broken-heart syndrome, to model default dependence in corporate bond-based CDOs. Li’s approach quickly became the industry standard (MacKenzie and Spears, 2014a: 405–406). The three major CRAs developed GCM-based software systems between 2001 and 2004 (MacKenzie and Spears, 2014a). By late 2003, subprime mortgages had become a favorite component of a new class of CDOs.

Regulation coevolved with financial innovation in ways that strengthened the incentive to create and hold MBS-based CDOs. Of particular importance was the decision by American regulatory authorities to introduce a ratings-based approach to bank capital risk weights in November 2001 (Friedman and Kraus, 2011; Miller, 2018). Under these new rules, AA and AAA securities carried a 20% risk weight, which incentivized banks to accumulate highly rated MBS CDOs. The new rules reflected the regulatory agencies’ embrace of the general philosophy of the risk revolution, as well as many of the specific types of risk-management techniques that banks had been developing. As Federal Reserve Board Chair Ben Bernanke (2006) noted, the ratings-based approach and the broader shift in regulatory philosophy it signaled had coevolved with changes in private institutions’ activities: “Banks and other market participants … made many of the key innovations,” while regulators “adapted and disseminated best practices” and established “guidelines that codify evolving practices.” He noted that this coevolution “has been particularly extensive in the field of bank capital regulation,” where new regulations “build on the risk-measurement and risk-management practices of the most sophisticated banking organizations” (Bernanke, 2006). ¹²

By 2004, market participants and the CRAs had embraced a single approach, the GCM, to construct, rate, and price CDOs (MacKenzie and Spears, 2014a, 2014b). Furthermore, because the CRAs relied upon the GCM, financial institutions tailored the ABS CDOs they constructed to conform to the CRA models (MacKenzie, 2011: 1786). The CRAs facilitated such tailoring by distributing their GCM-based software systems to the CDO makers and working with the investment banks in their construction. As the GCM became more prominent, network externalities reinforced its appeal. As one market participant related:

the GCM became performative … in that the act of me going out and saying, “this is a great valuation tool” meant everyone said, “we’ll use it.” Once everyone was using it, you have to use it as well, because it then becomes a good guide to prices. (MacKenzie and Spears, 2014a: 423)

As new regulations increased demand for CDOs, and as the industry’s embrace of the GCM made it easier to create and trade them, new issues of subprime mortgages skyrocketed, rising from US$100 billion in 2000 to a peak of US$600 billion per year in 2005 and 2006 (Financial Crisis Inquiry Commission, 2011: 70).

In short, CDOs emerged as a specific innovation within the context of a broader revolution, and spread through the financial system via positive feedback reinforced by the coevolution of private financial risk-management models and the regulatory regime. This risk revolution changed the possibility space of the (mortgage) finance system; events that had not been possible prior to 2003 suddenly became possible. Furthermore, because the transformation of finance altered the possibility space, it changed the probability distribution associated with the space. Some events became more likely and others became less likely. Indeed, this change in the probability distribution was precisely what regulators stressed to justify their embrace of the risk revolution: reducing the degree to which banks stored risk on their balance sheets made each individual bank and the system as a whole more stable.

Rather than distribute risk, however, the system concentrated risk in some of the system’s largest and most central institutions. Risk became more concentrated because as actors exploited information derived from the system’s past, they changed the way the system behaved (Boulding, 1981: 46). I call this the information–entropy cycle. ¹³ Over time, the accumulation of observations of the system enables actors to predict what events are more and less likely to occur. Actors then exploit this information in pursuit of specific goals. Yet, as they exploit information, they may alter behavior within the system and thereby reshape the probability distribution. As a consequence, information about system properties drawn from the past ceases to be a reliable guide for the present. Unfortunately, however, the information–entropy cycle often operates in a world of uncertainty in which actors cannot know that entropy is rising because they cannot directly observe the changing distribution. Instead, actors become aware of change when the system produces new information in the form of unexpected events (Blyth, 2010; Taleb and Pilpel, 2004).

The information–entropy cycle nicely captures the processes that led to the concentration of risk in the financial system after 2003. The GCM relies heavily on information about asset price correlations. Default dependence, the probability that the individual mortgages that constitute the pool from which a CDO is constructed will default simultaneously, is the key element in the transformation of a group of relatively high-risk subprime mortgages into a highly rated CDO tranche. As long as default dependence is low, only the investors in the equity tranche face any significant risk of loss. If default dependence is very high, however, then many defaults will occur simultaneously, likely eliminating the equity tranche and imposing losses on the mezzanine and senior tranches as well (Coval et al., 2009; MacKenzie and Spears, 2014b: 406). Having an accurate estimate of default dependence is thus critical to CDO performance.

Market participants, the CRAs, and the regulatory agencies assumed that historical information about the correlation between US home prices in different parts of the country provided a good estimate of contemporary default dependence in MBS CDOs. Historically, home prices in the various states had not been highly correlated. In 1980, for instance, the median state-to-state correlation of home prices stood at only .05, and home prices in as many as one-third of all state dyads were negatively correlated (Landier et al., 2017). Hence, if these correlations persisted, default dependence in a geographically diversified pool of loans would be quite low. Generally speaking, market participants and observers believed that this historical pattern continued to hold. As Case and Shiller (2003: 342) argued in 2003, “judging from the historical record, a nationwide drop in real housing prices is unlikely, and the drops in different cities are not likely to be synchronous: some will probably not occur for a number of years.” Such assumptions were broadly held. Moody’s, for instance, never incorporated into its GCM a stress scenario that included a nationwide decline in home prices (Financial Crisis Inquiry Commission, 2011: 120–121). Market participants thus exploited information about the system’s past behavior to assemble MBS CDOs from portfolios of geographically diversified subprime mortgages in the belief that such a pool had an intrinsically low default dependence (MacKenzie, 2011: 1816).

Yet, US home prices had not been highly correlated in the past because the regulatory environment had hindered the development of a national real estate market. Until the early 1990s, restrictions on interstate banking and branching sustained a segmented mortgage finance system in which the typical mortgage lender concentrated its activities within a single state or region. Segmented lending meant that mortgage lending cycles and the housing price movements that such cycles generated were influenced more by idiosyncratic developments within each state and were driven rather less by common national developments. Consequently, there was very low cross-regional and cross-state correlation in residential real estate prices. This regulatory environment changed after 1994, however, as a consequence of the Riegle-Neal Interstate Banking and Branching Efficiency Act. Ironically, perhaps, regulators encouraged nationwide banking in the wake of the savings and loan crisis in order to allow and encourage banks to diversify their balance sheets. As Lawrence Lindsay explained to the Financial Crisis Inquiry Commission:

If you had a regional … real estate downturn it took down the banks in that region along with it…. So we said to ourselves, “How on earth do we get around this problem?” And the answer was, “Let’s have a national securities market so we don’t have regional concentration.” (Financial Crisis Inquiry Commission, 2011: 43)

In another instance of coevolution, financial institutions altered their lending patterns in response to the 1994 regulatory change and, by doing so, increased the correlation between real estate price movements in different parts of the nation. Mortgage lending became more concentrated after 1994, with a smaller number of large banks lending across a wider geographic area. The emergence of structured finance accentuated this concentration. At the housing boom’s peak in 2004–2006, 80% (by value) of new issues were originated by 25 lenders (Keys et al., 2013: 151–153) while the share of mortgages originated by lenders whose business focused on a single local market had fallen to 4% (Loutskina and Strahan, 2011: 1447). As mortgage lending became more concentrated, it became more synchronized across the various states and thereby caused home price correlations to rise well above historical levels (see Kallberg et al., 2014; Landier et al., 2017). Kallberg et al. (2014), for instance, estimate that inter-regional correlations of home prices doubled between 1995 and 2005. Default dependence within the MBS CDOs thus increased significantly and, as a result, “the entire pool [of mortgages that made up a given CDO started] to behave like a single asset” (MacKenzie and Spears, 2014a: 404). Moreover, as many of the largest institutions had accumulated large portfolios of these essentially identical mortgage-backed CDOs, the entire banking system increasingly took on the characteristics of a single asset as well. In short, widespread efforts to exploit information drawn from the system’s past changed the system. Entropy began to rise — the information that actors employed to make decisions no longer described the system in which they were trying to manage risk.

Regulators and market participants did not know and could not have known that entropy was rising. In a world in which important characteristics of the system can change as an unintentional consequence of uncoordinated agent interaction, actors cannot know that the system has changed until the system generates information which suggests that something has changed. This is a pretty simple instance of what Taleb and Pilpel (2004) call “the unfortunate problem of the non-observability of the probability distribution.” In this case, the information required to prevent the crisis was the rising default dependence and the nationwide correlation in housing price movements that underpinned it. Regrettably, regulators and bankers could not observe beforehand that the probability that any downturn that occurred would be nationwide was rising and thus could not know that default dependence had increased. They could learn that the system had changed only when the system generated a nationwide decline in home prices. ¹⁴

A PECI perspective on the subprime crisis thus encourages us to focus on the mechanisms that changed the financial system over time and the consequences of this change for system performance. It stresses the impact of innovation, positive feedback, diffusion, and performativity within the financial system itself, as well as the coevolution of financial innovation and financial regulation. It stresses the non-ergodic nature of the financial system by exploring how these processes altered the financial system’s possibility space and the associated probability distribution. It encourages us to recognize that as actors change the system, they often unintentionally destroy information and thus operate in an environment of rising entropy in which their understanding of how the system works becomes increasingly inaccurate. Often, this cycle unfolds in a world of uncertainty in which actors cannot know that they have changed the system. Finally, the operation of this information–entropy cycle over time can generate large discontinuities in system performance. Furthermore, although I have focused on the subprime crisis, a PECI perspective has obvious application to the global financial system more broadly.

Conclusion: The epistemology of a complex system ¹⁵

At its most fundamental, the argument that I have presented constitutes a call for scholarship that studies the global political economy as a complex system. Such a call may seem puzzling and even unnecessary to many readers. I imagine that every scholar of the global political economy agrees that “a system” is the subject of our research. Moreover, most scholars, especially in the US, adhere to the proposition that understanding this system does not require us to study it as a system. Indeed, prevailing sentiment encourages us to “embrace partialism,” an epistemological strategy in which one learns about the system by studying its constituent parts and then fitting them together to create an image of the whole (Lake, 2011, 2013). The predominant perspective thus holds that studying the system is unnecessary and may even be counterproductive.

What the partialist perspective typically overlooks, however, is the question of what type of system it is that we study. As Kenneth Boulding (1987: 115) has pointed out, “the systems of the real world that we are trying to know about exist in great variety.” Boulding identifies 10 real-world systems, each of which possesses a unique theoretical logic. The underlying conceptual dimension across which these 10 systems vary is their degree of indeterminism. To quote Boulding (1987: 115–116):

[the] simplest systems are those that are stable over time, which can be investigated by simple observation and recording…. Next we have systems that change according to constant parameters, that are describable in terms of difference or differential equations of a given degree. These might be called “predictable systems” of which, of course, the classic example is celestial mechanics…. There are systems, however, that are inherently quite unpredictable, in which the failure to predict is not the deficiency in human knowledge, but an inherent and inescapable property of the system itself. These are systems in which information is an essential element…. As we move into social systems … indeterminism becomes even larger.

Knowing which of these types we study matters, for:

if we are to increase our understanding of [systems] and if our images of them in our minds are to contain a larger proportion of truth and a smaller proportion of error, then for each system we are investigating we have to search for an appropriate methodology. (Boulding, 1985: 17)

The recognition that systems come in varying forms, some of which are indeterminate and unpredictable, generates three clear implications for how we study IPE. The first implication concerns how we generate knowledge about the global political economy. The field must become more receptive to epistemologies that “embrace holism.” Embracing holism means studying the global political economy as a system. More generally, the logic of holism encourages us to take the ecosystem rather than the individual worker or firm as the unit of analysis, regardless of scale. For instance, study financial institutions in the context of the financial system that they inhabit and vary the scale of this ecosystem as appropriate and justifiable. Second, because the global political economy is a complex adaptive system in which time matters, IPE scholarship should draw more heavily from the historical sciences and less heavily from the experimental sciences. From a practical perspective, this means research that explains events rather than research that tests hypotheses in the search for regularities (Cleland, 2002, 2011). Finally, this means that scholars need to recognize that the knowledge about human social systems that we generate as scholars has a limited shelf life (Blyth, 2006). Such recognition may, but need not, have consequences for how we generate knowledge, but it will have implications for how we use it. In other words, social science research itself is not immune from the information–entropy cycle.