On investing

The future of work

Regarding this thought:

Regression algorithms of the kind that algorithmic trading may rely on for estimating security prices struggle with multi peaked distributions, eg if price may settle on 1–2 or 10–20, they will still project in range 2–10

Another modern phenomenon that could be attributed to algorithmic trading, rebalancing of holdings between securities often leverage the Kelly criterion for balancing risk reward. Many forget that rebalancing assumes ability to operate on time scales commensurate with the market.

There are a few ways to try and unpack the comments surrounding regression algorithms in context of multi peaked distributions.

Part of the reason this is an issue is that any consideration of an equity that needs to incorporate exposure to scenarios will be blind to scenarios that are aggregated into a single weighted projection of value.

The following is not based on extensive experience deriving my own algorithmic trading strategies, but I have some exposure to the field and even though most firms that deploy such tactics will likely have small or large deviations in their own methodologies, there will still be several common elements that one can expect to be similar between conventions.

From a high frequency trading standpoint, my impression is that majority of conventions are following pricing movement signals of an equity or surrounding industry metrics, which won’t typically account for long term dynamics that may arise from eg changes in paradigms. I get the sense that less precise long term trends that may get gradually incorporated into a price signal are more commonly coming from larger funds with large exposures and longer holding periods (the proverbial whales of the market). Unlike high frequency actions, such whales have a disadvantage in comparison to others in that their actions are more visible to the rest of the market, simply because in order to enter or exit from a huge position in some security, it becomes visible in price actions that aren’t otherwise supported by known causes (as may become apparent when say a particular equity starts to show signs of diverging from other participants in a segment of industry). I expect that if I was such a whale I would probably attempt to disguise such larger actions by way of conducting such trades in a manner resembling day trading or high frequency movements, but such forms of disguise should only work over longer time horizons in which the amount of transactions can be spread into less perceptible buckets. In conditions of more rapid changes to economic conditions, regulatory environments, or geopolitical matters, to name a few examples, there arise scenarios where a whale type investor who wishes to make material changes to their exposure will only be able to do so in a manner that becomes visible to the market.

If we consider an equity that has let’s just say three trading scenarios, one where it goes out of business (the zero case), one where it trades within some particular range indefinitely (that status quo case), and one where it reaches significant traction (the moon case), a probabilistic aggregation into a regression derived point forecast of the three derived based on prior conditions will likely underestimate the moon case and will otherwise obscure the probabilistic components into a single value. Typically those regression derivations that attempt to incorporate an uncertainty metric do little more that assigning a window of potential values in which they have an established certainty that the value will fall within — I typically see this as somewhere in the 90% or 95% range. So if an assessment is attempting to maintain some adherence to a risk tolerance, it is common that they will only make trades that align to maintaining their adherence to the eg 90% certainty case. (I saw this in the “conformal predictions” conventions described in several research venues but have seen in other frameworks as well, I think this mindset is fairly common.)

The problem with such trading strategies is that they completely omit crafting exposures to outlier events that on their own are unlikely within a fixed time period, but over long enough time periods they become increasingly likely to come about in some as yet unknown time period. In other words, just as exposure to some small risk compounds into a large probability over sufficient time periods, one can expect that a small probability of rewards will likewise compound into larger probabilities over sufficient time periods (assuming that such probability of rewards remains in effect in future time periods).

You will find in the market that most energy from participants who focus on outlier scenarios instead of maximal certainty investing will do so on the down side, by way of hedging against crashes in a particular equity or sometimes even markets in aggregate. I just don’t see first hand the papers or commentary from those focussing on outliers scenarios of equity liftoffs (outside of the field of venture capital).

I think what most people miss about the market, or about their philosophy of investing in general, is to compartmentalize their exposures to market pricing in comparison to other sources of cash flow. This probably comes about because people become accustomed to outsourcing their portfolio management to dedicated professionals. The problem with outsourcing such focus though is that not only are you giving up whatever portfolio management fees that may come into play, you are likewise giving up any emerging industry expertise that becomes available to those with skin in the game and who put sufficient focus into following narrow segments of industry for long enough time periods. This kind of industry expertise isn’t free of course, if you are just passively reading whatever garbage gets posted to online media you aren’t going to find much, the best way to gain such exposure to industry expertise is to seek to participate in the industry yourself. Build software, volunteer for professional associations, contribute to academia, approach potential clients, seek to add value and advance the field yourself. If you do this for long enough in the right industries, it kind of doesn’t matter what you are invested in, because even if some unforeseeable scenario changes the industry dynamics overnight, you have built in hedging by valuable connections, expertise, and employable skills.

I do expect though that in the age of AI, the value of such industry esoteric expertise will diminish in comparison to the simple and longstanding value of a strong professional network. People that you can connect to and establish new opportunities for sales or collaboration.

That is also why venues like twitter (aka “X”) for establishing a track record of public speech and leadership are invaluable. If anyone can look anything up in an online reference material, we are no longer marketing our knowledge, we are marketing our abililty to evaluate and assess new information signals and filter out the noise from the signal.

Put differently, if you are not an employee, you can still earn a right to become an industry participant in a few ways. Entrepreneurship is one, equity investing is another. The more you channel into such pursuits, the more reasonable is expectation for a real seat at the table.

Nicholas Teague

March 03, 2025