“In the short run, the market is a voting machine but in the long run it is a weighing machine.” - Benjamin Graham
Welcome back to Premium Income Investments! I hope you all had a great New Year and are looking forward towards 2023!
To kick off the New Year, I want to revisit something I have touched on a lot before, the idea of Market Efficiency. Why you ask?
Perhaps I’m a bit obsessed, but I think it has more to do with the fact that Market Efficiency is a central theme of this newsletter. We look for possible ways to outperform the markets, but with the general consensus that the public equity markets are completely efficient, our efforts may well be in vain.
And this is why today I want to reiterate my conviction that even today the public equity markets are not truly efficient.
At least, not all the time…
But before we continue, let me introduce you to InvestorSnippets. InvestorSnippets is a free daily newsletter which aggregates and summarizes a few important articles from big news outlets on stocks, ETFs, and markets. The newsletter can be digested in as little as two minutes if you read just the summaries, or if you want more information you can read the full article which is linked in the newsletter.
You can take a look at their most recent newsletter here if you want to see whether it’s your style.
Anyway! back to your regularly scheduled programming!
You see, the argument made by pure market efficiency proponents generally are along the lines of, “all past and present information has been incorporated into a particular stock’s price, therefore none of it is relevant for predicting future movements in price. The stock’s price movements follow what is called a ‘random walk’ and are entirely random.”
However, “random” is a somewhat misunderstood term. Nothing in our universe is truly “random” simply because they are driven by deterministic processes. (minus quantum mechanics) If enough information can be known about a situation, then in theory it can be predicted. In fact, there is evidence to suggest that machine learning can predict “random” number generators that have been implemented various different programming languages.
Cause and effect shape everything in such a way that the closest we can possibly get is “pseudo-randomness.”
Herein lies our key. Things are not actually random, they simply appear random. Which means if we could gather enough relevant data we may be able to predict future movements in the market.
The problem is, humans are not really that good at learning associations among large amounts of data.
Human judgment is far more limited than we think. We have a surprisingly restricted capacity to manage or interpret complex information. - David Faust in The Limits of Scientific Reasoning
In his book David Faust (director of Psychology at Rhode Island hospital and faculty member of the Brown University medical school) found that human judgements were consistently outperformed by simple actuarial models.
The point being, humans are just not always suited to finding the intricacies of a dataset. When presented with complex and vast amounts of data our decisions can become more subjective or reliant upon what “seems” correct.
Machines on the other hand, do not have this problem. When machines learn associations in data (via machine learning) they use more objective measures. More specifically speaking, neural networks use gradient descent which minimizes a loss function that relates to its data inputs.
Now I’m not necessarily saying that all market beating strategies require machine learning, case in point the Medallion Fund founded by James Simmons has been active and beating the market every year by a large margin long before machine learning was a thing. But, if we can use a model that has objective metrics for evaluating buying and selling conditions, there is a shot at beating the market.
So why aren’t machine learning models able to predict future stock prices from past data and technical indicators?
You might be referring to an article such as this. The author presents a machine learning model which really does a not so great job at predicting future stock prices.
You’ll notice that the predicted stock prices just look like a moving average indicator more or less. This is because the model was trained using past stock prices.
The problem with stock prices, is that they are the result of other data, they are not necessarily the means by which future stock prices depend upon.
On the other hand though…
If we use the data that influences the stock price, such as 10k reports, news, investor sentiment, institutional inflows and outflows, etc then we might actually have a model worth pursuing.
I’m not the only one who thinks this either…
Back in 2018 Two Sigma (one of the larger quant funds) released a challenge on Kaggle (a website for competitive machine learning) relating to predicting stock prices using news data.
If that’s not an indicator of the type of input data we should be using to train market predicting models, then I don’t know what is.
Anyway, that’s it for today’s article. If you liked it, please consider leaving a like or sharing it with a friend! It really helps me out!
Have a great week!
All content is for discussion, entertainment, and illustrative purposes only and should not be construed as professional financial advice, solicitation, or recommendation to buy or sell any securities, notwithstanding anything stated.
There are risks associated with investing in securities. Loss of principal is possible. Some high-risk investments may use leverage, which could accentuate losses. Foreign investing involves special risks, including a greater volatility and political, economic and currency risks and differences in accounting methods. Past performance is not a predictor of future investment performance.
Should you need such advice, consult a licensed financial advisor, legal advisor, or tax advisor.
All views expressed are personal opinion and are subject to change without responsibility to update views. No guarantee is given regarding the accuracy of information on this post
Thanks for the article, really good point and reminder. For all intents and purposes, due the limitations of the human mind to see patterns in large datasets, things may as well be random to us, but in the grand scheme of things, that's not the case.
I feel that the worrying part here over time, as with most things unfortunately, any edge achieved through technology or otherwise tends to benefit institutional investors first and more, leaving retail behind.