Where do I begin? It is enough to know that today’s post tries to pair up technology and finance in no particular structure or order, just a rambling of thoughts of which I have no time or desire to reorganise and structure. You have been warned.
There used to be a time when people needed traffic police at junctions to direct them, that is, until traffic lights came along. And before the barcode scanner, a time when the prices of groceries needed to be entered manually into the till. I’ve often wondered whether the traffic police and the person at the till ever envisioned that one day, things would be so very different.
It is my belief that we are seeing a turning point in how we ‘do’ finance. Let’s start with investing. No, let’s begin with the most famous investor during this time, Warren Buffett. With his own words, he described his process as a lot of reading, and back then, of accounts. The ideas and screening for possible investments seemed as he described it in the early days, very hands on and manual.
Fast forward sixty years later, the data that we use for screening is fed by Bloomberg directly into the models. The models in turn, are either based on relationships discovered by academic studies or in-house research. Filters make selections and eliminations of assets possible with a wider range, greater accuracy and in faster time. Much, much faster time and ease.
Apart from stock selection, other processes benefitting from automation and algorithms are: documentation, monitoring performance, asset allocation, back-testing, forecasting and of course, trading.
Currently, algorithms are useful for the execution of trades, but choosing the correct algorithm and the right trading strategy at the right time still falls on human traders. Perhaps one day there will be an algorithm that helps us choose which algorithm strategy we should deploy and how.
It is not only the ease of which traditional processes are carried out by using machines, but also the method by which we do it. Machine learning is done using an algorithm to ‘train’ the computer through probabilistic learning on the most efficient way to execute a task.
Exciting tasks such as bankruptcy prediction, pin-pointing fraud and sentiment classification are challenges that hopefully will be fully solved through machine learning. In addition, I will go so far as to claim that the rise of machine learning in investing spells the end of investors as we know them today.
[Segue on sentiment classification:
There is a problem with sentiment classification, thwarted expectation and keywords to be exact. For example, the sentiment of the sentence “This book should have been brilliant. The plot would have been gripping if not for the hero’s portrayal,” for a human reader should be obvious, i.e. the book sucks. However, a machine learning algorithm would pick up the words ‘brilliant’ and ‘gripping’ and give a different reading.
Company reports hence, should keep away from including negative words in their reports, even if reading the whole sentence would convey the positive sentiment because a machine reading would come to the opposite conclusion.
After having read company reports and conference calls for a while, I must note that restraint in reporting, nuance and enthusiasm are not easily detected, if not impossible to be picked up through machine learning with the current technology.]
Automation of investment processes is portrayed to be a black box and increasing risks as a result. This may be true in the short run, but I have faith that we will tinker, adapt and build more complete models resulting in better risk management. What is now the cause of increased volatility especially during tail events may prove to be the stabiliser of our financial industry in the long run.
Why is this so? The average human cannot calculate as fast as a computer could. Equally, the average human cannot reason logically as fast as a machine learning algorithm could, chess games being an example. Humans also fall foul to behavioural weaknesses that compel them to take loss-inducing actions that could otherwise have been avoided.
Think of the financial technology as a house renovation, where you have to tear down walls and build new ones, dust everywhere and an inconvenience to your daily routine, but once it is completed, and if completed well, will improve your way of life and increase the space of your ‘house’.
Why risk is a concern is because the advancement of risk management program has not kept up with the returns side. Uncovering anomalies and chasing alpha have been given way more attention than the finessing of risk management algorithm, although this area too will catch up.
The real danger for the long run in my opinion, is when AIs monitor other AIs, resulting in a whole series of ‘reflected’ trades. This could result in the divergence from the fundamentals of asset valuation. Turning around that heavy tanker once launched would prove a huge challenge to all.
Some question why after many decades, there have been no visible improvements, no flying cars and no Star Trek replicators. But we have been improving. We have been improving something that is not visible. We have been trying to improve how we trade and allocate capital via technology.
Should we pin so much hope on machine learning and finance? I’m not so sure. This tool if done correctly can take us further. On the other hand, just like the Hollywood producers who churn the same type of movies over and over because they ‘work’, machine learnt investing might go for the same type of investment and investing procedures or trends because they too, ‘work’. To advance humankind, we sometimes need more than just a black box, we need the wizard who thinks outside the box, who will invest in something that others would think will never work.