In Models We Trust: An Investigation of Confidence in Algorithmic Trading

On 15 September 2008, the day Lehman Brothers filed for bankruptcy, chair of the US Federal Reserve Ben Bernanke gave a talk at the Economic Club of New York. In his talk, Bernanke spoke about the burgeoning crisis in the US financial system and said that ‘the root of the problem is a loss of confidence by investors and the public in the strength of key financial institutions and markets’. He went on to claim that the crisis would only come to an end ‘when comprehensive responses by political and financial leaders restore that trust’. To Bernanke, the crisis was essentially a collapse of confidence. The concept of confidence is hard to pin down, which might be why the role of confidence in maintaining the financial system remains an understudied topic in modern economics and modern economic sociology. Keynes pointed to this issue arguing that confidence is a term ‘practical men’ pay close and anxious attention to, while economists only discuss it ‘in general terms’ (Keynes, 1936: 148–9). Swedberg’s work has made up for some of the scholarly negligence of the role of confidence in stabilising and sustaining the financial system (Swedberg, 2010, 2012b, 2012a, 2013). In this paper, we contend that Swedberg’s theory of confidence applies not only on the level of inter-connected financial institutions, but also on the level of individual market actors. Moreover, we argue that it provides a good starting point for devising a theory of confidence that accounts for the socio-material entanglements of human actors, models, algorithms, and data that characterise contemporary algorithmic trading. Since confidence is conspicuous by its absence, we study instances of loss and lack of confidence and discuss implications on trading and investment practices. We draw on two sets of empirical material: one that considers the relation between investor and money manager and the other the relation between the investment firm and their trading algorithms. First, we examine letters from hedge fund managers sent to their investors during the 2007 Quant Meltdown, which was a weeklong slump in the US equities markets that almost exclusively hit model-driven hedge funds. As a second source of empirical material, we draw on interviews with market participants working in data- and model-driven investment management firms. We study how quants, traders and portfolio managers handle underperforming models and who and/or what they blame when models are performing poorly. We argue that confidence is instrumental when traders and portfolio managers consider whether to override or retire a model. Informed by the empirical analyses, we attempt to devise a theory of confidence that accounts for the interplay between different market participants and the algorithms that execute, informs, and sometimes determines trading. Despite the increasing quantification and automation of the financial industry, which has rendered almost every decision and action measurable, calculable and classifiable, the study demonstrates that confidence remains crucial in financial markets.