The study of the brain is a very fascinating subject – its learning process astounds me, and its stumbles confound me.
DeepMind just defeated the Go World Champion, a game considered too complex for computers until now. The machines are coming. If you had to program a machine to pick stocks, what would you teach it?
I know quant funds exist. I am referring here to picking stocks with conviction sizing, where a machine is able to say Idea A is so much better than Idea B. Here I am reflecting on how you will assess Idea A is better and what rules you
are implicitly or explicitly using.
I am a keen student of epistemology, the process of knowledge transfer. Early in my career, I worked on documenting how credit analysis was done at a global bank in London and New York, with a view to replicate it in India. These efforts were successful because we could recreate substantially all the conditions needed for contextual understanding. The ingredients in this case were company financials (3 years), an understanding of IFRS and US GAAP Accounting, an understanding of historical and present trends in the industry (3rd party reports), and a basic understanding of the economic and cultural trends. The last was the most nuanced. It was easier to do this for the
UK and US, as culturally the bridge was not far to cross. American and British shows and movies were the predominant part of the cultural diet for the English-speaking population. News flow was easy to get, and the leading
financial newspapers of those countries (The Wall Street Journal; The Financial Times) were available both in hard copy and on the Internet.
The pace of knowledge transfer is also important to understand. Hard copy newspapers stopped being relevant for me in the Lehman crisis. The state of the crisis, related bailouts, positioning of the Congress, statements by the Fed – all of these did not lend itself well to being frozen in print for one day. On most days, news is slow. But when news is fast, the newspaper failed to deliver. Since then, I am in favour of online news sources, even for these reputed newspapers. It allows continuous updates, quick fixes if any error is present, and the ability to read reactions across the spectrum, because of the online comments section. I have learnt a lot more by reading informed comments, deepening my understanding of the subject many times more than if I had just read the article in print.
Can experienced fund managers share ideas well? I have found this to be more difficult in practise than in theory. Each manager comes with his/her set of experiences, and sector and thematic preferences. They may have their own views of alignment. For example, I do not see high insider ownership as a strong indicator of alignment. I mention earlier the example we are currently seeing in the portfolio. When an individual is very rich, and has wealth outside of the company, they have other options. When a firm is intrinsically linked to a key founder, it becomes difficult for anyone else to confidently say they can run it better, and there is no effective counter to a cheap take-private offer. This is very often seen in Asia, where scions of the business family follow in the footsteps of the patriarch. The families regularly make money in each cycle by listing or privatising companies from their empire, keeping core companies private, while partially listing satellite companies. Unless you are large enough in the company to have a blocking stake – over 10% for most countries to block a delisting, or over 25% to block a majority vote – you should not be in this segment of the market. And they do take an inter-generational time horizon for their decisions, so you cannot take positions unless you can take a 10-15-year view. You cannot create value of a different timescale than how management seeks to achieve it. It might happen by timing, but that is not something that can be planned and prepared for.
Accounting, since we need to understand the numbers. Then about incentives and agency issues (management vs. owner).
An understanding of the balance between labour and capital. Political history, to understand government policies and direction.
Market sentiment, crowd psychology and individual biases.
And you could add industry stacks of knowledge, more knowledge about understanding people and motivation, and judging character.
Some say people have intuition, machines don’t. But perhaps intuition is nothing but a set of rules expressed really fast. My quick reaction to a news item is based on predefined rules that I may have trouble articulating one by one, but they do exist. If fund managers regularly train and teach apprentices over the years, it is not hard to program a machine to learn and understand. We teach rules. Then exceptions to the rules. Then other rules which may apply. Then the priorities of the rules. Then exceptions when those priorities are different and so on. Given that humans make mistakes, a machine which picks stocks with some acceptable level of mistakes is not far away. Many already exist in rudimentary form. A stock screen is a machine. If you clean up the database, the stock screen performs better. If you improve the screen to treat different sectors differently, it gets better. (e.g. high debt may be acceptable for an utility while it increases risk to an unacceptable level for a commodity producer). We interact with machines all the time, we train them and improve them. We teach younger/older mentees and colleagues, and we can both learn if we listen properly to push backs.
I guess the most important thing is learning how to learn, and to keep learning. Is high insider ownership a good sign? It depends.