Mark Fleming-Williams of The Alternative Data Podcast and I discuss some history of quantitative finance, specifically the tools and ideas I have seen influence the space over the last two decades. I had a lot of fun talking with Mark, and am a huge fan of his podcasts. If you haven’t yet listened, I recommend looking through his archive.
Hope you enjoy listening as much as I enjoyed the conversation!
If there is one thing that matters more than anything else in systematic trading, it’s arguably the speed at which you convert an idea into action. There are many points of failure along the way, from discovering alpha to implementing it, but the process itself can’t be one of them.
I had the opportunity to be part of a great panel discussion a few weeks ago on the topic of the need for speed in alpha generation. What was discussed was the absolute breakneck pace at which research requirements, especially compute-centric artificial intelligence and machine learning algorithms, have pushed the technical limitations of what is possible. New algorithms and faster machines, including cloud compute options, are also on the table, but a major factor in speed is often overlooked. The speed I am referring to is the speed of research, to go from initial idea to realizing a trade.
This process may seem simple at first, but in a systematic environment there are myriad steps that need to happen to take you from idea to PnL. From data on-boarding, exploration, and cleaning, to feature extraction and signal generation, to backtesting and portfolio construction. Each of these pre-trade steps in turn have multiple steps - which often means multiple points of failure. Maybe bad timestamps. Maybe missing symbology. Maybe it just doesn’t even fit in memory. Maybe a database that slows to a crawl at the wrong moment, or a networks goes down. How you handle these steps can take weeks in ad-hoc systems. If the data is at all novel, it’s highly unlikely anyone has solved your exact problem before. Every new idea seems to have to start at square one. A proper system to manage this at both scale (size of data and number of sources) as well as at velocity of ideas, is a rare and elusive animal. Every day lost, every compute cycle wasted is calculable lost alpha. Your research pipeline directly impacts your success.
To be successful, you will forever need more and better ideas — but you need to be able to test and trade them faster than your competitors. The only edge that doesn’t go away is the edge you build into your process. How to generate ideas. How to test ideas. How to use ideas. These are an often overlooked, and undervalued, part of quant trading. This is the hard work that isn’t talked about, but it is the effort expended here that matters most.
Copyright 2020. Lemnica Corp.