When trying to find exploitable market patterns that one could trade as a retail trader, one of the first questions is: What trading frequencies to look at? Monthly? Weekly? Daily? Or intraday anywhere between 5 seconds to 1 hour? With limited time available for conducting research at all of these timescales, this becomes an important question to answer. I and others have observed that there seems to be a simple relationship between trading frequency and amount of effort needed to find a profitable strategy that is purely quantitative and has acceptable risk. In short:
The lower (=slower) the frequency you want to trade at, the ‘smarter’ your profitable strategy needs to be.
As an example, one could look at the (very) high frequency end of the spectrum, where marketmaking strategies based on really very simple mathematics can be very profitable, if you manage to be close enough to the market center.
Taking a big jump into the daily frequency realm, it is becoming much harder to find quantitative strategies that are profitable while still being based on rather simple mathematics.
Trading in weekly and monthly intervals, using simple quantitative methods or ‘technical’ indicators only is a very good recipe for disaster.
So, assuming for a moment that this relationship is indeed true and also considering that we can and want to use sophisticated machine learning techniques in our trading strategies, we could start with a weekly frequency window and work our way towards higher frequencies.
Weekly trading does not have to be automated at all and can be done from any web-based brokerage interface. We could develop a bag of strategies, using publicly available historical data in combination with our favourite learning algorithm to find tradeable market patterns and then execute the strategy manually. At this scale, all the effort should go into finding and fine-tuning the quantitative strategy and very little thought needs to be put into trade execution. Trade automation effort: 0%. Strategy smartness required: 100%
Daily trading should be automated, unless you can really dedicate a fixed portion of your day to monitoring the markets and executing trades. Integrating machine learning algorithms with automated daily trading is not a trivial tasks, but it can be done. Trade automation effort: 20%, Strategy smartness required: 80%
On intraday timescales, ranging from minutes and seconds to sub-seconds, the effort you will have to undertake to automate your trades can lie anywhere in the range between 20% and 90%. Fortunately the smaller the timescale becomes the ‘dumber’ your strategy can be, but ‘dumb’ is of course a relative concept here. Trade automation effort: 80%, Strategy smartness required: 20%