I questioned whether I should be running a more mechanical system back in January. At the time I decided I wasn’t quite ready to hand over the controls to my portfolio just yet, but would set up a mechanical benchmark portfolio to track my performance against. It’s too early to reach any conclusions but the early signs for the mechanical portfolio are good – it’s been on quite a tear this year so far.
With all the data and information technology available to us these days, in one sense I see the challenge of investing as about finding the optimal blend of human and machine. You want to find the best balance between the high level nuanced judgment and pattern recognition of a human and the emotionless discipline (and brute calculating ability) of a machine. Trying to identify the aspects of my strategy where a more systematic or algorithmic approach may work better than human judgment, and vice versa, seems like an important thing to do.
I’m happy with how this balance works for identifying high quality businesses for my watchlist. The computer is clearly better placed to do the quantitative screening, while I’m confident that my judgment in making a qualitative assessment of whether a business is high quality adds a lot of value.
I still feel I haven’t yet found the best balance for timing trades (i.e. what to buy and sell from the watchlist and when). Over time I am becoming more and more drawn to idea that this process should be largely systematic. After all, the whole game is to exploit the behavioural errors made by others while not making them yourself. A systematic approach seems better suited for this than human judgment – the human may be prone to subconsciously make the same mistakes!
This is basically the idea behind my mechanical momentum benchmark portfolio. It only trades in the same high quality shares as my actual portfolio. The difference is it trades by simply swapping the lowest momentum holding in my portfolio for the highest momentum holding on my watchlist once a week. This rotation is ‘partial’ in that it only swaps one share rather than rebalancing the whole portfolio. The idea is to consistently maximise exposure to momentum, while limiting the amount of trading (and trading costs) to some degree. It will endure periods of underperformance as momentum strategies do, but over the longer term I think I can be confident that it should outperform.
So far the mechanical portfolio has outpaced my actual portfolio significantly. It’s 22% up since inception on January 25th, versus 13% for my actual portfolio. This doesn’t account for the trading costs incurred in the mechanical portfolio, but even accounting for these it would still be significantly ahead. It’s far too short a time period to read much into, but I have been reflecting on why the mechanical system has been doing better than my actual portfolio and if there is anything that I might learn from this.
Lessons from the mechanical portfolio
Stop losses vs periodic rotation
My current strategy relies quite heavily on stop losses in deciding when to sell. A concern I’ve been having with this is that stop losses mean you are always selling at a short term low point in the share price. This can be fine if your investments are making big moves up with low volatility before finally tailing off. However, when prices are more volatile, stop losses often make you sell at the worst points. Because of this, I think a strategy which relies too heavily on stop losses when deciding when to sell faces a significant weakness. It may be sensible to use them, but you don’t want this to be the only way you sell!
I have noticed that the mechanical portfolio may be benefiting from using periodic rotation instead of stop losses. Periodic rotation seems to provide better discipline against overtrading in volatile markets. While stop losses might take me out of lots of shares at the same time, periodic rotation forces this process to slow down. This allows more scope for some shares to recover from temporary weakness, while still eliminating at a measured pace those that suffer more prolonged weakness. Given my tendency to overtrade, I think the discipline from periodic rotation could be beneficial.
I’m also drawn to periodic rotation because it seems more consistent and direct ‘top-down’ way of implementing the ideas behind my strategy. It would ensure that my portfolio as a whole maximises its exposure to the factors I want to be exposed to. Periodic rotation forces you to directly consider the qualities of different investments relative to one another. On the other hand, using stop losses seems to be more aligned with thinking about each individual investment on a case by case basis. Stop losses are absolute rather than relative and provide the discipline to effectively manage the risk of each individual trade. This has its own advantages, but I’m coming to the view that using stop losses effectively may require a somewhat different mentality and approach to the one I employ.
Discipline against biases
It’s easy to fall into the trap of thinking that because you are aware of certain behavioural biases that you are in some way immune to them. I’ve written previously about the behavioural biases that give rise to momentum. Observing the performance of the mechanical benchmark portfolio has made me concerned that I may be falling into some of these ways of thinking on occasion.
The most striking thing about the mechanical portfolio is that it has successfully captured a large part of pretty much all the major positive price moves from the shares on my watchlist. While it hasn’t always been invested in a share before a price move has started, it has been consistently effective at investing pretty soon afterwards. While I’ve been well aware of these moves happening, I haven’t been nearly as good at capitalising on them in my actual portfolio. I’ve tried to be more selective and avoided those opportunities that looked too expensive or where I felt I was too late, but more often not this layer of judgment seems to be hindering rather than helping me. To put it bluntly, the more mechanical approach seems to have been much better at harvesting the excess returns from momentum than I have.
So what to change?
A purely mechanical approach is very effective at focusing systematically on a single important factor (i.e. momentum in this context). While I think the discipline this provides is hugely important, there are some downsides. I don’t think the scope for human judgment to add value is excluded entirely:
- There is still scope for value to be added by concentrating to a greater extent in investments where I have higher conviction. There are some cases where I am significantly more confident that the business is drastically mispriced given its quality. Given that my highest conviction investments tend to also have high momentum, the discretion to hold significantly larger positions in my most concentrated investments doesn’t look like it would interfere very much with an otherwise mechanical strategy.
- I’ve noticed that there seems to be a much higher degree of correlation between the investments in the mechanical portfolio – consequently it tends to be hit much harder on down days and has a more volatile performance overall. I think this means there is probably a role for some intervention to ensure there is sufficient diversification across sectors and geographies.
- It is well-known that momentum doesn’t work well in certain market conditions. ‘Momentum crashes’ can be pretty severe. I don’t think these are a major cause for concern, as the conditions where momentum starts performing relatively badly I think are identifiable i.e. at the low point of a market crash. A drastic shift in strategy would be warranted at these points. There is a good article on Stockopedia about this.
Adapting the mechanical approach to allow for these elements of discretion seems fairly straightforward. However, I’m still not ready to fully embrace this quite yet. The idea of fully leaving the controls to a mechanical momentum system is rather scary, particularly at this point in the cycle! There could well be issues with it I have not yet realised. I think a sensible stepping stone for now is for me to follow the mechanical system more closely as a benchmark guide. As a start, I’m going to shift towards trading more periodically (though I’m thinking once every two weeks at most) rather than relying on stop losses.
Hi, thanks, really interesting read. Out of interest, how have you gone about setting up a mechanical trading trading system?
I’m not sure I’ve understood your question. It’s a benchmark rather than an actual ‘live’ portfolio. I make dummy trades based on actual prices to track it (with a portfolio set up in Stockopedia). By ‘mechanical’ I just mean systematic or rules-based with no discretion. Hopefully that clarifies?
Many thank for a very interesting post. How have you defined the momentum indicator used in your mechanical system? (in your previous post you suggest using either the six month price momentum or the Stockopedia momentum rank). To further complicate things six month price momentum could either mean the Relative Strength (Stockopedia definition) or the % price change. Which are you using for your benchmark portfolio?
That is a good question. I have actually set up several benchmark portfolios using different momentum indicators to compare across them. I didn’t elaborate on this in this post as performance was pretty similar across them.
I’ve used the Momentum rank, 1y price change, 6m price change, % vs 52 week high. I preferred not to use RS as I’m investing across multiple geographies and I also wanted the system to allocate to the geography with the best momentum.
The Momentum rank is quite different to the others as it also includes whether there are recent earnings beats and analyst upgrades. I like this idea a lot in principle, but I’m finding that in practice the data quality (especially for small caps) can be a bit patchy. While it seems to work v well on average, for certain shares there are sometimes anomalous readings.
Of the others, % vs 52 week high is interesting as it also introduces an aspect of low volatility as well as momentum (you can have low difference to 52 week high if the price is flat as well as if it is rising…). I like this idea in principle but I think you would ideally want to use a blend of this as well as the absolute price change.