Final Report Summary - EPIFM (Evaluation Practices in Financial Markets)
The EPIFM (Evaluation Practices in Financial Markets) project investigated how professional practitioners in financial markets evaluate – attempt to work out the monetary worth of – financial instruments such as shares and government bonds. The EPIFM team are sociologists, an anthropologist and political scientists, not economists, and we sought to understand the broad range of factors that influence evaluation practices, factors that we found to include: technological change; organisational context; the relationships among organisations (including their relative power); regulation; and whether and how monetary evaluation interacts with other ways of thinking about 'value'. Although we applied quantitative methods where appropriate, much of the data collection we conducted was via interviews with practitioners, supplemented by direct observation and primary-source historical research.
The first phase of the project consisted of a broad-scope empirical investigation of the topic, focussing primarily on the work of investment-management firms, including their portfolio managers and analysts. What we found was that this work was shaped profoundly by relationships among firms and individuals that can be characterised as the 'investment chain'. Most investment flows through an often extended sequence of intermediaries, and the relations among these intermediaries are both enabling (at the most basic, they make it possible for investment management to be pursued on the giant scale it currently is) and constraining (intermediaries form crucial, and sometimes deeply sceptical, audiences for each other's performances of successful, professional, financial selves). This analysis is presented at length in our joint book, Chains of Finance: How Investment Management is Shaped (Oxford University Press, 2017).
The second phase of the project was a set of nine more specialised case studies, covering topics such as the shift from evaluation directly by human beings to evaluation by algorithms (especially the ultrafast algorithms of high-frequency trading or HFT), the historical evolution of electronic trading (including the role in that evolution of moral concerns), the continuing role even in today's electronic markets of lay people as traders, how investors' organisational arrangements and evaluation practices influence governments' capacity to borrow (and the terms at which they can borrow), and the roles played by important categories of mathematical model in evaluation practices.
Two examples from our case studies give a flavour of what we did and what we found. Principal Investigator Donald MacKenzie and postdoc Taylor Spears studied the development of, reception of, and uses of a family of models known as 'Gaussian copulas'. From the 1990s onwards, these models became employed widely by banks, credit rating agencies, etc. to evaluate complex 'credit derivatives' such as 'collateralised debt obligations' or CDOs. We found that specialists – including some of those who had contributed to the development of the Gaussian copula family – often disliked the very models to which they had contributed. The models, however, had become entrenched as the 'market standard', and it was organisationally very difficult for a bank, for example, not to use them. Remarkably, indeed, we found that this entrenchment continued after the 2007-8 global crisis in which Gaussian copula models were implicated: the organisational costs of abandoning them were too high. MacKenzie and Spears published these findings in two articles in the leading specialist journal, Social Studies of Science, in 2014.
The second example concerns how HFT algorithms trading US shares make their decisions to buy or sell those shares (or cancel their existing orders to buy or sell). MacKenzie's extensive interviewing of practitioners of HFT discovered that four main classes of 'signal' are employed by these algorithms (a 'signal' is a pattern of data used by algorithms in these decisions). He then examined the histories of these classes, finding that in three of the four cases the existence of the 'signals' was the result of struggles that had a broadly 'political economy' nature, involving traders, exchanges, regulators and sometimes politicians. This analysis was published in 2018 in the world's leading sociology journal, the American Journal of Sociology.
The first phase of the project consisted of a broad-scope empirical investigation of the topic, focussing primarily on the work of investment-management firms, including their portfolio managers and analysts. What we found was that this work was shaped profoundly by relationships among firms and individuals that can be characterised as the 'investment chain'. Most investment flows through an often extended sequence of intermediaries, and the relations among these intermediaries are both enabling (at the most basic, they make it possible for investment management to be pursued on the giant scale it currently is) and constraining (intermediaries form crucial, and sometimes deeply sceptical, audiences for each other's performances of successful, professional, financial selves). This analysis is presented at length in our joint book, Chains of Finance: How Investment Management is Shaped (Oxford University Press, 2017).
The second phase of the project was a set of nine more specialised case studies, covering topics such as the shift from evaluation directly by human beings to evaluation by algorithms (especially the ultrafast algorithms of high-frequency trading or HFT), the historical evolution of electronic trading (including the role in that evolution of moral concerns), the continuing role even in today's electronic markets of lay people as traders, how investors' organisational arrangements and evaluation practices influence governments' capacity to borrow (and the terms at which they can borrow), and the roles played by important categories of mathematical model in evaluation practices.
Two examples from our case studies give a flavour of what we did and what we found. Principal Investigator Donald MacKenzie and postdoc Taylor Spears studied the development of, reception of, and uses of a family of models known as 'Gaussian copulas'. From the 1990s onwards, these models became employed widely by banks, credit rating agencies, etc. to evaluate complex 'credit derivatives' such as 'collateralised debt obligations' or CDOs. We found that specialists – including some of those who had contributed to the development of the Gaussian copula family – often disliked the very models to which they had contributed. The models, however, had become entrenched as the 'market standard', and it was organisationally very difficult for a bank, for example, not to use them. Remarkably, indeed, we found that this entrenchment continued after the 2007-8 global crisis in which Gaussian copula models were implicated: the organisational costs of abandoning them were too high. MacKenzie and Spears published these findings in two articles in the leading specialist journal, Social Studies of Science, in 2014.
The second example concerns how HFT algorithms trading US shares make their decisions to buy or sell those shares (or cancel their existing orders to buy or sell). MacKenzie's extensive interviewing of practitioners of HFT discovered that four main classes of 'signal' are employed by these algorithms (a 'signal' is a pattern of data used by algorithms in these decisions). He then examined the histories of these classes, finding that in three of the four cases the existence of the 'signals' was the result of struggles that had a broadly 'political economy' nature, involving traders, exchanges, regulators and sometimes politicians. This analysis was published in 2018 in the world's leading sociology journal, the American Journal of Sociology.