The Evolution of Valuation and related challenges: Nonlinearity, Learning and Interpretability

Abstract

We explore aspects of the ongoing evolution of valuation in the sell side (investment banking) and in the buy side (hedge funds, asset managers) following the 2008 crisis. We highlight how nonlinear effects that were previously negligible took over and changed valuation in the sell side introducing barely adequate valuation adjustments.

These adjustments account for closeout cash flows at default of a counterparty, funding costs, capital costs, effects of margins, but in reality would not really be separable in additive adjustments except under special assumptions. The full valuation would require non-linear quantitative tools that are not currently adopted by most banks. The current methodology may lead to possible double counting and valuation errors. Some of the nonlinear effects are not transparent and bring back the contrast between price and value. We further highlight how the emphasis moved from complex products with simple costs/risks to simple products with complex costs/risks. For the buy side, the basic paradigms still consider linearity and normality as the norm, and often adopt the standard deviation as a risk measure. We show an example of how this may have aggravated 2018 losses in short volatility positions and how these might have been reduced by using Gaussian mixtures for returns and Expected Shortfall as a risk measure. However, fully embracing nonlinearity would mean resorting to Deep Learning, as was done by some firms for alternative risk premia, which raises the fundamental problem of interpretability. We briefly mention some initial applications of interpretability in finance and emphasize the need for more work in this area.

The seminar will take place online via Zoom