How cross-position contradictions destroy portfolio returns

A book of ten positions is rarely a book of ten independent bets. It is usually a book of ten positions that share three or four underlying assumptions about the world — and at least one of those assumptions is being contradicted by a position elsewhere in the same book.

This is not a diversification failure in the conventional sense. The positions can be in different sectors, different geographies, different durations, even on opposite sides of the same trade. The contradiction is at the level of the reasoning, not the exposure. You can be long ASML on "semis capex is resilient through any plausible recession" and simultaneously long CVX on "capex-led oil demand will recover into the next downturn." The first thesis says capex holds up; the second thesis says capex returns from a trough. Both can't be the version of the world you're betting on.

The damage compounds quietly. In benign markets, neither thesis is stress-tested and both make money, so the contradiction is invisible. When the regime shifts — say capex collapses — the ASML thesis dies for the right reasons and the CVX thesis dies for the wrong reasons, both at the same time, and you discover you were running twice the exposure to the capex factor you thought you were.

The conventional defenses don't catch this. Sector limits don't catch it because the positions are in different sectors. Factor models don't catch it because the contradiction is at the level of a specific causal claim, not a standard risk factor. Position-level risk reviews don't catch it because each position is internally coherent — you have to read two theses against each other to see the conflict.

The fix is structural. Every thesis has to be written with its load-bearing premise made explicit, and the portfolio has to be readable as a set of premises rather than a set of tickers. "I am long X because of premise P1." "I am short Y because of premise P2." If P1 and P2 are inconsistent, you have a problem that no position-level review will surface.

In practice, almost no investor does this, because doing it requires extracting the premise from each thesis and then comparing premises pairwise across the book. That's an O(n²) reading task that humans don't do voluntarily on a book of more than five positions. It's the kind of task that has to be done at the level of the tool, not the discipline.

The returns you lose to cross-position contradictions don't show up in your P&L attribution. They show up as a portfolio that's quieter than it should be in the good times — because contradicting theses are partially canceling each other — and louder than it should be in the bad times, because they stop canceling at exactly the wrong moment.