I like lists. I like making plans. Recently, I built an entire website for a one-week holiday in Ireland using GPT sites. It worked really, really well.
At university, I studied everything from archaeology to chemical engineering, physics and seismology. I speak two languages equally poorly and can barely write in either. English included.
I approach life as a problem that can be solved by finding the right set of steps. I suppose that’s why I’m an optimist.
Reading this, you might imagine a naturally organised, disciplined person, with nary a file or diary entry out of place.
That could not be further from the truth. I am organically lazy. If I think I can jump from step two to step ten, I’ll try. I’ll often fail, then try again.
A repeatable investment process is something I have to build and maintain. It does not come free with the personality.
Because of this, not despite it, I keep coming back to process and rules. More importantly, I look for ways to make my lazy, creative and frequently distracted self pay attention long enough to act.
This article is one example of how I try to do that, using real investments and market events.
Trigger warning: there is some mafs involved. Sorry, but ye cannae avoid that, I’m afraid. I’ve tried to keep it simple, but some complexity and the occasional big word are necessary.
Please comment if anything is unclear. Explaining something simply is surprisingly difficult when it has become second nature after twenty years. What looks obvious to me may leave you scratching your head.
One thing a useful process should do is draw your attention to a change you might otherwise overlook, especially when overlooking it feels quite pleasant.
In October 2025, several investments began to bother me for a slightly awkward reason. They were all doing rather well.
Metals, nuclear power, rare earths and quantum computing had different investment cases. But their increasingly large moves together raised a practical question: did the portfolio still contain as much diversification as its collection of stories suggested?
Before getting to the numbers, it is worth explaining how those different ideas can end up in the same portfolio.
How those ideas get into a portfolio
At AlphaInvestors, our model portfolios start with a working view of growth, inflation, policy and the cost of money. Then come the specific questions: where is capital being spent, what is in short supply, and which changes in technology or demand matter?
That view informs four broad allocations. Beta supplies general market exposure. Diversifiers bring different sources of return or protection. Themes express more specific investment cases. Cash controls overall exposure and leaves room to act.
Rising electricity demand can lead to nuclear generation. Supply insecurity can lead to rare earths. Monetary risk can support metals. Quantum needs its own argument about commercial progress, financing and how much future success is already in the price.
The theme is only the beginning of the investment work. Who gets paid? When do the cash flows arrive? How much capital is needed first? What return is left at today’s valuation? A sound economic argument can still lead to the wrong company, or the right company at an absurd price.
The final step is to size the positions together. Metals do not become a reliable hedge simply because they sit under “diversifiers”. A nuclear holding can share risks with a quantum business despite producing something entirely different. Broad equity funds may already contain some of the exposures added through themes.
Those relationships help determine how much of each investment the portfolio can sensibly carry.
13 October: a very good day
On 13 October 2025, Rigetti rose about 25%, Oklo 16%, the rare-earth pair 20%, and metals 3%. All four groups were having a day in the strongest tenth of their own previous sixty-session experience.
Seeing all of that green would usually make for a pleasant afternoon. The question was whether the portfolio was being paid for four separate investment cases, or increasingly for one common enthusiasm.
There was a credible shared catalyst. That day, JPMorgan announced a ten-year, $1.5 trillion financing and investment initiative whose coverage included critical minerals, nuclear energy and quantum computing. About $1 trillion was already planned; direct equity and venture investment was up to $10 billion. The headline was not a cheque to these companies. But the prospect of better access to capital offered a reason for otherwise different businesses to respond together.
Speculation could amplify that response. Neither explanation required every underlying business to have changed overnight.
For the investigation, the four groups receive equal weight in a research basket. The paired holdings split their group’s allocation. These are deliberately simple, aggressive weights, not a model portfolio. The newly launched MEME fund appears in the price story but not in the longer risk calculations: it had almost no history and already included quantum holdings.
What followed made the question more pressing. The four individual shares and MEME reached their October closing highs between the 13th and 15th.
Metals turned in the same short window, with silver’s October closing high on the 16th and gold’s on the 20th. Their subsequent losses were much smaller.
Similar-looking charts are a reason to investigate. They are not enough to decide how much to sell.
Were the positions moving together more often?
Correlation gives the first answer. It measures whether investments tend to have their stronger and weaker days together, rather than whether they merely happened to make similar gains over a longer stretch.
That distinction matters to anyone holding more than one position. If one investment’s weaker days are often offset by another’s stronger days, the combination is steadier. If they increasingly have bad days together, the benefit shrinks. No extra purchase is required for the portfolio to become riskier.
Across the six pairs in this four-group basket, average twenty-session correlation rose from 0.11 on 10 October to 0.33 on the 13th and 0.49 on the 21st.
The holdings were providing less diversification just when several had become much more exciting to own.
Could this just be a statistical trick? The result did not depend on choosing twenty days: ten- and forty-session estimates also rose. Removing broad market, technology, Treasury and dollar exposures did not remove the relationship either. After those controls, average correlation was about 0.13 before the episode and 0.44 in the following period.
There is an important weakness, though. A calculation confined to the six-day peak window depends heavily on the enormous 13 October rally. Remove that day and most of its apparent correlation disappears. The longer windows are more informative. Even those describe a small historical sample, not a permanent law about these investments.
The shared behaviour did not simply disappear after prices fell. Average twenty-session correlation was still about 0.59 on 31 March 2026. That makes this more useful as a study of portfolio risk than as evidence of a tidy, two-week bubble.
How much diversification was actually left?
Four equal allocations look balanced on a spreadsheet. That does not tell us whether the holdings are responding to four different influences.
The Herfindahl–Hirschman Index, or HHI, measures concentration. Here it is applied to the statistical patterns in the holdings’ movements, rather than the money in each holding. It squares each pattern’s share and adds the results, so a dominant pattern counts heavily.
Before doing this, each group’s returns are adjusted for its usual volatility. Otherwise quantum would dominate the exercise simply because it tends to swing much more than metals.
Its inverse gives an effective number of movement patterns: how many equally important patterns would produce the same concentration. A higher number means movement is more evenly spread across the patterns.
Even before the sharp rise in correlation, the basket was closer to three patterns than four. By the 21st, it was closer to two, and the dominant pattern pulled all four groups in the same direction.
“2.2 patterns” does not mean exactly 2.2 independent investments. Nor does HHI identify the economic cause. It summarises the same correlation information, rather than casting a second, independent vote for a bubble.
Direction matters. Investments that reliably offset one another can also have highly patterned movements. Here, the growing common pattern pulled the holdings together, while the portfolio owned them all on the same side.
The practical consequence is to review their combined size. As diversification does less work, combined risk moves closer to the sum of the holdings’ weighted standalone risks. Seeing them all surge together should at least make you wonder how they might fall.
Were the joint moves unusually extreme?
Back to the mafs. Correlation and HHI describe relationships over time. Turbulence asks how unusual today’s combination of moves is, given how these investments normally behave together.
It considers both the size of the individual moves and their usual relationships. A large fall in a volatile stock is one thing. A large fall alongside an investment expected to cushion it may be much more unusual. The same test can flag an enormous shared rally.
On 13 October, all four groups jumped sharply. On the 21st, all four fell sharply. Each day was in the strongest or weakest tenth, respectively, of each group’s previous sixty-session experience.
A useful check asks how much of the signal came from the size of the moves, and how much came from the way they occurred together. Much of the turbulence came from the sheer scale of the individual returns, rather than an unprecedented breakdown in correlations.
That would strengthen my suspicion that enthusiasm had become excessive and that several supposedly separate bets were leaning on a common driver. It would not establish that every investment had become exactly the same trade. But it would certainly prompt a review of their combined size.
The important question was now: how much more risk would these positions carry if the recent behaviour persisted?
A slow view for normal conditions, a fast view for change
In my institutional portfolios, I used a slow view and a fast view of risk. I carry that approach into my models and personal portfolios. Here is the simple version.
The slow model uses weekly observations with relatively slow decay. Older information gradually loses influence, so the assessment changes over months rather than jumping around every few sessions. In a stable regime, that smoothness is useful.
The fast model uses daily observations with much faster decay. Recent days carry more influence, allowing it to react to the structure visible over the last week.
The faster view becomes particularly useful when there is a reason to believe the structure has changed. Previously unrelated investments may start moving together. Stocks and bonds may stop offsetting each other. A relationship that made the portfolio manageable last month may no longer be helping.
The slow estimate still includes plenty of the old behaviour. The fast estimate asks what risk looks like if the latest behaviour deserves more weight.
After the 13 October rally, the research basket’s fast risk estimate was 59% above its 10 October level. By the 21st it was roughly 80% higher, while the slower estimate had risen about 4%. At my old asset-management firms, complex portfolios led us to build our own risk infrastructure. In this example, lights could have been flashing on the desk while a slow system upstairs barely moved.
Volatility here means the estimated size of day-to-day fluctuations. It is not a maximum possible loss.
A fast model can also respond to a disturbance that disappears. Selling into one of those episodes can leave money behind when the rally resumes. That is why the model and the risk policy need to be chosen before the manager starts negotiating with an uncomfortable result.
The constant question: when do I take profits?
Some version of this comes up repeatedly: “Brendan, the Shell trade has done really well. When do I take profits?”
The answer starts with the original investment case, the return still available and the size of the position. There are three reasons to reduce it. Our quantitative approach is the third, but it can also nudge you to revisit the other two.
1. The remaining return may no longer justify the risk. Suppose a medium-term investment was bought at 100 because 120 looked plausible. If it reaches 118 in six weeks and nothing changes the target, only about 1.7% remains before dividends and costs. The question is whether that remaining return is worth the downside, not how satisfying the original profit feels. The numbers are illustrative, not a valuation of Shell.
2. The position may simply have become too large. A holding intended to occupy 5% of the portfolio can grow to 8% during a strong run. Rebalancing towards the original weight brings exposure back towards the amount originally chosen. It does not require concluding that the company has become a bad investment.
3. Its contribution to portfolio risk may have risen. The holding might still occupy 5%, but swing more violently, move more closely with the rest of the book, or both. Rebalancing to an unchanged capital weight would not necessarily solve that problem.
The third is the problem this episode illustrates. Keep the risk budget fixed and change exposure when measured risk changes. That can mean taking profits after a rally or cutting a loser. The rule responds to risk, not whether the position has made money.
A pre-agreed allowance makes the decision less dependent on how large a profit feels, or how uncomfortable the latest trading day has become.
So how much should come out?
Suppose the allowance is fixed at the basket’s 10 October risk level. The same mix now carries about 80% more estimated risk per dollar invested. Leaving the exposure unchanged accepts a larger risk budget without consciously choosing one.
To restore the original allowance, the sizing snapshot says retain a little over half the exposure and move about 45% to cash.
Even holding individual volatilities unchanged, the observed correlation change alone increased estimated basket risk by about 18%, enough to imply a reduction of roughly 15%. The larger fast-model adjustment reflects the full change in both volatility and relationships.
Correlation, HHI and turbulence explain why the portfolio deserves attention. The estimated risk of the holdings, measured against the chosen allowance, determines how much exposure fits. They are not three independent alarms that must all agree before a sale can happen.
This is where the process earns its place. The risk allowance is agreed before the exciting day arrives. Once the holdings exceed it, the calculation gives a starting point for the reduction. There is still judgment involved in choosing the model, setting the allowance and deciding what to keep. But the amount of risk being taken is no longer left to whichever version of the investor happens to be at the desk that afternoon.
For a live Alpha portfolio, the calculation must include the rest of the book: beta, diversifiers, other themes and cash. A 45% cut to this stand-alone research basket does not mean a 45% cut to a real portfolio.
And after bringing the risk back inside the limit, the remaining capital still has to be allocated. Which investment has the best prospective return? Which thesis is weakest? Where is the same exposure already hiding elsewhere in the portfolio? A proportional cut is a transparent starting point, not a substitute for security selection.
What would cutting risk actually have changed?
Now follow the whole run-up and reversal. Start with $20,000 at the 1 October 2025 close, divided equally among the four research groups, and track both accounts through 14 November. One leaves every share untouched. The other checks risk after each close and, when necessary, reduces the shares at the next session’s recorded opening prices.
The allowance is fixed using information available at the 1 October close. The calculation tracks changing weights, cash and overnight gaps, with trading friction of 0.10% of sale proceeds, no cash interest and no buying back in. This is a retrospective sizing example, not actual Alpha trades or a complete strategy.
The rule produces three reductions. The first comes on 3 October, before the mid-October episode. Later reductions respond to the portfolio still held, including the cash already raised.
The diagnostics explain the episode; the orders here come from the volatility budget. They do not wait for correlation, concentration and turbulence all to cross a threshold.
The untouched basket rises by about 55%, then gives back all of that gain and more. The reduction account also participates in the rally, but carries less exposure into the reversal. It finishes with about $3,914 more capital: $21,843 versus $17,929, after assumed trading costs.
There is a visible price for that protection. At the 14 October closing high, leaving the shares alone is about $2,079 ahead. The rule sacrifices part of a rally that is still working. The reduced account also subsequently falls almost 25% from its own closing high. Smaller positions help; they do not make speculative holdings safe.
Making only the first cut ends at about $19,207, rather than $21,843. Continuing to review the changed portfolio therefore helps in this case. But it does not trigger another sale on 22 October: existing cash and changed holdings keep risk below the ceiling. A stand-alone “cut 45%” calculation is not an instruction to keep selling.
The job is to notice when holdings carry more risk than intended, calculate a proportionate response, and keep checking the resulting portfolio. The dated orders and full ledger are in the methods.
The takeaway: different investment stories can become one much larger risk. A useful process notices the change and turns it into a decision that could turn loss into profit.
What this does, and does not, establish
During a strong run, each profitable theme can seem to confirm its own story, even as the holdings become more dependent on the same market behaviour. A slow model can miss that change. Emotionally, the bigger danger is feeling clever and adding risk because it all looks so easy. This I have been guilty of many times!
Tracking correlation, concentration and unusual joint moves helps draw attention to the change. Translating it into risk makes the observation actionable: how much exposure still fits, and how much needs to come out?
That is the sort of process I find useful. It gives a distracted investor something specific to pay attention to, and a decision to make while there is still time to make it deliberately. More importantly, it stops me doing stupid things like adding to a winner when the market has already increased the bet for me.
The more I reflect on investing and life, the more I think avoiding stupid things improves the outcome. As someone who has done a helluva lot of stupid stuff, expect more on that topic.
Let’s end on this image, “Vague Intellectual Pleasure”, taken from Annie Besant and Charles Leadbeater’s Thought-Forms (1901), a Victorian book of mysticism that looked at visualizing feelings and thoughts as images, from the wonderful Public Domain Review.
Hopefully, it describes your own thought form if you read until the end…