Machine learning transforms the landscape of quantitative finance

History of Quantitative Finance

We can describe quantitative finance as the use of mathematical and statistical methods in the financial context, with applications in areas like pricing, trading, risk management among others.

While simple mathematical models for pricing were created soon after the introduction of the first derivative contracts in the 1700s, quantitative finance really expanded during the 20th century. It benefited from advances in mathematics (like the development of the theory of stochastic processes) and the rapid improvement of existing computational tools.

A first milestone can be placed in 1900, when Louis Bachelier proposed in his PhD thesis to model the time series of prices for financial products by means of a Brownian motion. This connection was disregarded for about half a century. But, it blossomed in the 1970s with the appearance of papers like “The pricing of options and corporate liabilities” by Fischer Black and Myron Scholes, and “On the pricing of corporate debt: the risk structure of interest rates” by Robert Merton. These authors predated a true cascade of academic works on different areas of application. More importantly, it had a huge practical impact, opening the door for the introduction of many new kinds of derivative contracts, advanced regulation, and the creation of systematic methods for trading.

An interesting detailed historical account can be found in A brief history of quantitative finance [1].

New Landscape of Quantitative Finance

Machine learning (ML) techniques exploit the large availability of data and the great improvements in computation power to provide alternative ways of solving analysis tasks with minimal external input

These characteristics have convinced financial companies to increasingly adopt advanced ML techniques to win the edge in the market competition. For instance, ML is been used for predicting market trends and portfolio selection; for pricing and hedging exotic derivatives as in DeepHedging; for calculating initial value adjustments as in Deep MVA; and for synthetic data generation as explained for example synthetic data given by JP morgan website.

ML can be also applied successfully to accelerate repetitive tasks: a perfect example of this is JPMorgan Chase’s COIN, which stands for contract intelligence. It takes about (360,000) hours for humans to review (12,000) commercial credit agreements, while the COIN analyzes legal documents and contracts using image recognition software, completing its task in a matter of seconds (see JPMorgan software does in seconds what took lawyers 360,000 hours). Other examples are discussed in Machine Learning in finance.

All in all, ML has shown great potential in either increasing modelling power or reducing the time and resources needed to accomplish some tasks. For this reason, ML is now an indispensable tool for any professional in the “quant” space.

References

[1] Cesa, M., 2017. A brief history of quantitative finance. Probab Uncertain Quant Risk 2, 6, s41546-017-0018–3.