In early 2018 JP Morgan released one of the most comprehensive studies ever (almost 300 pages!) on Machine Learning and Big Data pertaining to Financial Services. The study highlights many of the methods associated with machine learning aren’t actually new concepts- on a somewhat rudimentary basis some investors dealt with alternative datasets as early as the 1950s, when the founder of Walmart, Sam Walton, used small aircraft to fly over car parks to count cars in order to calculate real estate investments. The study was carried out by the quantitative investing and derivatives strategy team in New York (led by Dr Marko Kolanovic) and essentially confirms how machine learning will play a key role in future operations of financial markets. Nothing new here you might think? To save you time we have broken down the report into 7 key sections: 1. The cleansing of data is as important as its analysis Datasets these days are so vast and are constantly growing fed by a variety of sources, namely by individuals (product reviews, social media posts, search trends…), businesses (credit card data, transaction data…) and IoT / sensor data (aircraft locations, satellite imagery, map data…). This data needs to be cleansed or prepared in such a way which then can be properly acquired and analysed by quant researchers and data scientists who then apply their methodologies to this ‘alpha content’ in developing actionable insights. 2. Machines trump humans in making short and medium-term trading decisions According to JP Morgan the majority of high frequency trades are now carried out by machine learning algorithms: “machines have the ability to quickly analyse news feeds, tweets, process earning statements, scrape websites and trade on these instantaneously”. This will decrease the need for equity long short managers, fundamental analysts and macro investors. However, it is not looking all that negative for human beings: “machines will likely not do well in assessing regime changes (market turning points) and forecasts which involve interpreting more complicated human responses such as those of politicians and central bankers, understanding client positioning, or anticipating crowding”. 3. There are different varieties of machine learning Everyone is talking about machine learning these days, but there are several different variations to which this umbrella term can refer, for example supervised/ unsupervised learning, deep and reinforcement learning. The main difference between supervised/ unsupervised learning is that the latter attempts to work out the layout of data and understand the main drivers behind it. Deep learning is all about analysing a trend through means of neural networks and reinforcement learning exists to push algorithms in the direction of maximising trading profit. 4. Deep learning systems will carry out roles which are hard for humans to define but easy to carry out In a nutshell deep learning is the way of artificially recreating human intelligence. It is very adept in processed vast sets of unstructured data sets, for example is processing satellite imagery to count the number of trees planted in a certain area. In the same way that in the human brain each neuron passes on a message to other neurons which then work out a weighted average of the input, so is deep learning also applies a relative weighting of such inputs guided by prior experience. 5. Being an out-and-out Data Scientist/ Machine Learning Expert isn’t necessarily a requirement, but being a strong quant and programmer is JP Morgan remarks that a strong data scientist is essentially the equivalent to a quantitative researcher. “It is much easier for a quant researcher to change the format/size of a dataset, and employ better statistical and Machine Learning tools, than for an IT expert, Silicon Valley entrepreneur, or academic to learn how to design a viable trading strategy”. Since the majority of machine learning methodology are already coded (mostly in R), it is not necessarily a requirement to have strong expertise in machine learning, as you are essentially just applying existing models. 6. Support functions will also need to get to grips with Big Data as well The study is particularly critical of hiring managers and Headhunters who fail to distinguish between a general conversation around artificial intelligence and then how to formulate a tangible strategy as to how big data analytics can be applied. Compliance Teams within banks should also be able to understand models and critique them when possible to make sure that its content is properly anonymised (heard of GDPR?) and doesn’t contain and personal information. 7. These are the main data analysis packages and coding languages you need to know According to JP Morgan, if you only have time to learn one coding language this should be R, however, Python, C++ and Java also can be used by machine learning packages (for example keras with Python, Mallet with Java and Open CV in C++).