Proxy Always Falls Short of the Thing Itself
The habitat of market seams: existence is predictable, location is not
Rule Summary
The Rule: Any system that uses a quantifiable metric to approximate a real but invisible system leaves a seam behind. That seam is the cost of making the system run. It shifts as the rule tightens, but it never closes.
The mechanism underneath is simple. The underlying thing cannot be measured directly, so a system approximates it through observable proxies. Under normal conditions the proxy is good enough and the gap stays hidden. During a regime shift the proxy fails, and the gap widens to its maximum.
There is a boundary, and it matters more than the rule itself: this does not apply to anyone who reads “I know the seam exists” as “I can find the seam.” Existence is predictable. Location cannot be reduced to a formula.
That leaves three paths. You can build a disciplined search process and pair it with the judgment to separate real seams from fake ones. You can admit you lack that capability and fall back on passive market pricing. Or you can do what most people do and treat the proxy as truth. Each path has a cost.
One guardrail before we start. A seam must exist, but that does not mean it can be arbitraged directly, and it certainly does not guarantee that the thing you are looking at this time is the seam.
A threshold that wants to measure understanding but can only measure records
Start with an example I find worth turning over every time it comes up. Most investors have run into some version of it, and almost none stop to think about it.
Take the suitability requirements for leveraged and inverse ETFs. Daily rebalancing and the return drift it produces over long holding periods are features ordinary stocks do not have, so brokers and regulators set a higher qualification bar than for plain equities. One common gate is options-trading approval, often granted on the basis of your prior trading history. The intent is clear enough. Someone with a track record presumably understands what daily reset and leverage will eat, and is less likely to mistake 2x leverage for 2x returns.
The problem is that what the gate can check and what it wants to check are two different things.
It wants to measure whether a person understands risk. What it can actually measure is whether that person has accumulated enough trades in their account. Between those two sits a seam. The bar genuinely stops people with no trading history at all. Whether having a trading history means you actually understand risk is a different question, and the gap between the two never quite closes. Understanding and record-keeping were never the same thing.
And this is only one regulatory threshold and its design trade-off. What I want to get at is the structure underneath it, the reason this kind of gap is almost impossible to eliminate.
The rule uses a quantifiable threshold to approximate an unquantifiable quality
Pry that threshold open and a much larger structure sits underneath.
Any rule, once it has to govern enough people, cannot assess each person’s true quality one by one. No regulator can sit down with a vast population of investors and confirm, individually, whether you understand risk. It can only grab a proxy that is observable and auditable, then push that into the gap. Trade count stands in for risk awareness, net worth for investment maturity. A diploma or a certification does the same job for professional competence.
The advantage of a proxy is that it functions. The cost is that it never equals the underlying.
This is where the seam comes from. It is not a loophole some lazy official left behind. It is the price of making the rule run at all. To make a rule enforceable, you have to compress complex reality into a simple metric. The moment you compress, the parts of the underlying that the metric cannot capture spill out and become a seam. Records can be bought. Understanding cannot. A quantifiable gate stops people who can’t be bothered to assemble a record. It does not stop people who are willing to assemble a record while understanding nothing.
In engineering terms, this is an observability problem. The state you want to control cannot be measured directly, so you infer it from visible outputs. The error between the output and the true state is the cost of making the system controllable at all.
So the friction is the feature. Rules constrain the rule-followers, not the rule-gamers. That old saying is describing the same thing. What a rule stops is the person who acts according to the record. The person who treats the record as a puzzle to be solved walks right through. And the seam will not be filled, because the only way to fill it is to replace the proxy with the underlying itself, and the underlying cannot be measured. All you can do is raise the bar and tighten it, but that just relocates the seam. It does not close it.
The market uses visible proxies to approximate invisible reality
Move the same structure up one level and you get what the market does every day.
Price has to price risk, but the underlying of risk is invisible. No one can directly observe real panic, the real probability of default, or the future path of contagion. The market can only observe through proxies. Implied volatility on options approximates panic; VIX is the market’s standard read on it. Bond prices approximate default probability, which the market reads off the credit spread. Past correlations approximate future contagion.
Under normal conditions these proxies are good enough. On ordinary days VIX tracks real sentiment, spreads track real credit, correlations stay roughly stable. The gap is small, small enough to ignore, so everyone starts using the proxy as if it were the underlying.
The real trouble shows up the moment the regime shifts.
When the structure itself is changing, the link between proxy and underlying breaks. At the peak of panic, the price of hedging instruments can distort as liquidity dries up, and VIX stops measuring panic itself. March 2020 is the clean case. The index spiked to record highs, but part of that move was forced hedging and a dash for cash rather than fear alone, and in the same weeks the Treasury market, supposedly the deepest in the world, briefly seized up. When default risk is highest, spreads may fail to print because no one dares to quote. When you most need diversification, assets that are normally negatively correlated crash together, and the whole historical-correlation framework fails at once. The moment the proxy fails is exactly the moment the seam is widest, and it is also the moment when the largest number of people are still clinging to their ordinary proxies, convinced they can read the situation.
I will not draw any present-day conclusion here, and I am not implying any direction. The proxy is least reliable precisely when you need it most. The tool was built for normal conditions, and a regime shift is exactly the case it cannot handle.
The crowd uses narrative to approximate structure
One more level up, and you reach consensus.
When a group of people forms a shared view of a complex world, they rarely each read the structure for themselves. Mostly they share a story. “After the crisis comes the rebirth.” “This time is different.” “What’s fallen this far has to revert to the mean.” These are narratives. The function of a narrative is to compress a structure too complex for anyone to fully grasp into a single line that travels through a dinner-table conversation or a 30-second pitch.
To travel, a narrative has to simplify, and what gets simplified away is the seam.
Japan’s lost three decades is a clean example, useful because the narrative flipped twice. In the late 1980s, “Japan is going to buy the whole world” was the consensus, a story that compressed an extraordinarily complex structure into a simple sense of direction. The Nikkei peaked at the end of 1989 and broke the following year, but the narrative kept coasting on inertia long after the structure had turned. By the 2000s the consensus had inverted into “Japan is permanently deflationary and will never recover,” again compressing the complex into the simple. Both seams grew inside exactly the details each story sanded off to travel well.
Here is the counterintuitive part. You might think the stronger the consensus, the smaller the seam. The opposite is true. The stronger the consensus, the deeper the seam is hidden, because fewer people question the narrative and almost no one checks the gap between underlying and story. You tell yourself that so many people can’t all be wrong. But a large crowd only means the story spread widely. It does not mean the story sits close to the structure.
The habitat of the seam: existence is predictable, location is not
Stack the three layers and the shared skeleton emerges.
The regulatory layer, the market layer, and the consensus layer are all doing the same thing: managing a complex underlying with a simplified proxy. As long as that move happens, there is a gap between proxy and underlying, and the seam must exist. That part is predictable. I can say with confidence that every complex system has a seam, because every complex system is forced to simplify.
Now the reversal that matters most.
The existence of the seam is predictable. The location of the seam cannot be reduced to a formula.
These are very different claims. Knowing there is game in the forest is one thing. Knowing which tree this particular animal is hiding behind is another capability entirely. The first is logical inference. The second is judgment. Can you write a formula in advance that takes in market data and spits out “the seam is here this time”? No. Because the moment such a formula exists and enough people use it, the formula itself changes the behavior of participants. The original seam is either closed or squeezed into a corner the formula cannot see. This is the inevitability of reflexivity: a positioning method, once it spreads, rewrites the very thing it was built to position.
I spent fifteen years in chip verification, so this structure is very familiar to me. In verification we use coverage to approximate “is the design actually correct.” Driving coverage to 100% does not mean there are no bugs. It only means every state you tested passed. What can be automated is enumerating the uncovered states one by one. What cannot be automated is judging whether a given uncovered state is a real hole or a don’t-care that will never occur. The first goes to the tool. The second can only go to a person.
Market seams work the same way. The part that can be searched systematically belongs to discipline, and quantitative methods and AI do a great deal at this layer: they cast the candidate net wider and sweep the corners more thoroughly than a tired human ever could. But once the search returns a pile of candidates, deciding which one is a real seam and which is a trap someone laid is judgment, and that still rests on a person for now.
So the source of alpha is accurate identification, not faster search.
Judgment cannot be automated, and the limit is structural. What can be automated is, by definition, what can be written as a rule, and a seam is precisely the part that no rule describes. That is why it exists at all. You can hand the machine wider search and steadier discipline at lower emotional cost. What stays with you is recognizing which of its candidates actually sits close to the underlying, and that recognition is the one component in this system you cannot outsource.
Who this rule does not apply to
With the mechanism laid out, the boundary needs stating. The easiest misreading is to take “a seam must exist” and turn it straight into “the market is full of money, go grab it.” So the boundary comes first.
This rule is especially dangerous for one kind of person: the one who takes “the seam exists” and reads it directly as “I can find the seam.” Existence and locatability are two different things. Knowing the seam is surely there, and having the ability to locate where this particular seam sits, are separated by a great deal of research and discipline, and by a mind willing to admit “I might be wrong this time.” Without that last stretch, “a seam must exist” becomes nothing but fuel for overconfidence.
It also does not apply to anyone without a disciplined search process. Search without discipline turns up mostly noise, or bait someone deliberately put out.
There is one more boundary. In a market segment efficient and liquid enough, the seam may be too small to be worth the cost of searching for it. Judging whether a seam is worth chasing is itself part of the judgment. Seeing a seam is different from knowing that its expected return covers the cost of search and error.
For an allocator, three paths
The first is to build the search process and the identification judgment yourself. You scan for seams systematically, then use judgment to throw out the fakes, and most of the work is the throwing out. The hit rate is low by nature and the time commitment is long. The part most people underestimate is the emotional cost of sitting through long stretches that produce nothing. This path fits an allocator who has real research capacity and the stomach to hold the line through a run of low hit rates. It is expensive, and for most people it is the wrong choice.
For everyone else, the honest move is to admit you cannot locate the seam and take market pricing instead. Index or outsource. You give up active return and sit permanently at the average error of the proxy, which is a rational price to pay if you have positive cash flow and no need to beat the market to live on.
The third path is the dangerous one, and it is the one most people take without noticing they have taken it. You treat the proxy as the underlying. You trust the threshold, trust VIX, trust whatever narrative the consensus is running this quarter, and you forget there was ever a gap at all. The cost stays hidden until a regime shift, and then it arrives all at once: the proxies fail together, and you are the last to find out that every instrument in your hand has stopped reading the thing it was built to read.
The seam will always be there, because simplification will always be there. Tighten one rule and it reappears under the next. Reading where it sits this time is judgment. Searching for it with discipline is what gives that judgment something to work on. You need both. Lose either one and the whole thing fails, and neither half is something you can hand to a machine to finish.
Disclaimer
This article reflects my personal investment philosophy. It is not investment advice. Make your own informed decisions.
Miyama Capital manages proprietary capital only and does not solicit external investors.
Legal Disclaimer
This memo represents the author’s personal views on macroeconomic conditions, interest rate environments, and asset allocation as of the date of writing. It does not constitute a solicitation, recommendation, or guarantee regarding the purchase or sale of any security, fund, bond, or other financial instrument. Investing involves risk; bond prices, interest rates, foreign exchange rates, and economic/policy conditions may materially affect asset values. Scenarios and instruments discussed may become inapplicable as market conditions change. Readers who make investment decisions based on this memo do so at their own risk, and the author accepts no liability for any gains or losses arising from the use or citation of this material.
Kuan H. Wang Founder & CIO, Miyama Capital