Position sizing and the investment committee

Size is the only lever entirely in the institution's hands — and a committee paper is decidable only once what is unknown is written in it.

This lesson has had no expert review. It was written for this platform and against the evidence it cites; nobody has gone through it independently.

Learning objectives

You can derive a position cap from loss tolerance rather than from expected return.
You can identify when a correlation assumption is an assumption rather than a measurement.
You can check a committee paper for the four parts without which it is not decidable.

Check your prior knowledge

Answer these for yourself before reading on. Wherever you hesitate is where this lesson pays off.

What loss in a single position could your mandate absorb without the mandate itself coming into question?
Where do the correlations in your risk report come from?
What does the “exit” section of your papers say?

Core concept

Fact

Size is the lever you have

A protocol's risk cannot be changed from outside: neither the code, nor the governance, nor the dependencies respond to an institution being invested. The only adjustable quantity is how much of it is carried. So position size starts from loss tolerance rather than from expected return — the return decides whether to invest at all, the tolerance decides how much.

Assumption

Two assumptions every size rests on

Every cap rests on two assumptions: how much is lost on failure (the loss ratio) and that several positions do not fail at once. Both are as a rule estimated, not measured. Naming them does not soften the number — it makes it checkable, because afterwards one can see which of the two turned out wrong.

Uncertainty

Correlation from a short history is an estimate

Many on-chain positions have a price history measured in months rather than decades, and mostly from one market regime. A correlation computed from it is arithmetically correct and weak as a statement about the next stress event — precisely because dependencies then take effect that were invisible in the calm period. The sturdier route runs through the shared components rather than the time series.

Interpretation

What makes a paper decidable

A committee can only decide if four things are in the paper: the question being voted on; the evidence with source and date; what is expressly unknown; and the condition on which the position is unwound. Without the third, ignorance is recorded as agreement. Without the fourth, the exit later becomes a fresh debate under worse conditions.

Definitions

Loss tolerance: The loss a mandate can absorb without the mandate itself coming into question.
Loss given failure: The share of a position lost in the assumed failure case.
Exit condition: The observable condition, fixed in advance, on whose occurrence a position is unwound.

Model

Fix the loss tolerance for this case — before any look at returns
Assume the loss ratio, justify it, date it
Compute the cap
Check shared triggers and apply the cap to the group
Fix the exit condition before the position is opened

The order of a sizing decision — Expected return does not appear in this order — it decides whether, not how much.

Formulas

Cap from loss tolerance

max_size = loss_tolerance / loss_given_failure

max_size: Share of the portfolio the position may occupy at most
loss_tolerance: Bearable portfolio loss arising from this one position
loss_given_failure: Assumed share of the position lost in the failure case

Limit: The calculation assumes exactly one position fails. Where several positions share a trigger, the cap applies to the sum of those positions rather than to each — otherwise the same tolerance is spent several times over.

Worked example

The same tolerance, two assumptions

Loss tolerance per single case: 0.5 percent of the portfolio
Assumption A: total loss of the position in the failure case
Assumption B: 30 percent loss, since part stays collateralized
Shared trigger: two further positions depend on the same oracle

Under A: 0.5% / 1.0 = 0.5% cap. Under B: 0.5% / 0.3 = 1.67%. Because of the shared trigger, though, the respective cap applies to the sum of the three positions.

The cap differs by more than a factor of three — purely because of one estimated quantity.

Reading: That makes the loss ratio the most important number in the paper, to be included with a rationale and a date. Assumption B additionally requires evidence that the collateral would be realizable in the failure case at all — otherwise it is merely the more comfortable number without a basis.

Retrieval

Nothing — each position is its own case.The same tolerance is spent three times; with a shared trigger the cap applies to the sum of the three.The loss tolerance would have to be tripled per position.

The expected return with a range.The explicit list of what is unknown.The history of past returns.

Exercise on real data

Use the guiding questions to work out which figures would be needed to state an exit condition observably.

Dimension 6: liquidity →

Pick out the metrics this platform actually serves — an exit condition should rest on those.

Metric catalog →

Application

Write the resolution for a first on-chain position so it can be voted on without a follow-up question.

Related case studies

Institutional reading

Asset management: How many of your existing positions have a written exit condition?
Bank: Where would the cap live in your limit system — per position or per trigger?
Advisory: Which of the assumptions would you expressly flag to a client as an estimate?

Metrics in this lesson

Key takeaways

Position size follows from loss tolerance, not from expected return.
With a shared trigger the cap applies to the group, otherwise the same tolerance is spent several times.
Without a named gap in knowledge and an exit condition, a paper is not decidable.

Evidence

EVD-2026-0001
Electronic Markets (Springer) — A multivocal literature review of decentralized finance: Current knowledge and future research avenues