Research · Essay
Building a Non-Default Monthly Outlook
How we describe a month without ever naming a price target.
By Forecasting ✦ ✦3 minutes’ reading
In Brief
- The monthly outlook begins with questions, not targets. We select roughly ten whose answers would most change what we do.
- Each question is forecast independently, compared with an external probability, and published with its consensus gap and our confidence.
- The combined structure is compressed into a strategy state: regime, risk budget, conviction. Every outlook is scored the following month.
Most market outlooks follow the same template. A house view is stated as a set of targets — a level for the index, a yield for the ten-year, a range for the currency — and a narrative is written to support them. The targets are rarely revisited. When they are, it is usually to explain why they were directionally right.
Beginning this month, we will publish a monthly outlook of a different kind. It contains no targets. It consists of about ten binary questions, our probability for each, the external probability for each, and a short explanation of where and why they differ. This note describes how it is built.
§ IChoosing the questions
The hardest part of the process is the first one. A month can be described by thousands of possible questions; most are irrelevant, many are redundant, and some are unanswerable. The decomposition stage generates several hundred candidates from the current evidence base and then ranks them by a single criterion: how much would knowing the answer change our allocation?
Questions with high value are kept. Questions whose resolution would change nothing are discarded, however interesting. Of the remainder, highly correlated questions are merged, and anything that cannot be resolved against a public source on a fixed date is rewritten or dropped.
| Candidate question | Decision | Reason |
|---|---|---|
| Will the index rise in June? | Rejected | Outcome, not a driver. Answering it tells us nothing about why. |
| Will sentiment deteriorate? | Rejected | No unambiguous resolution source. |
| Will the central bank surprise markets? | Rewritten | Replaced with a specific, swaps-resolvable threshold. |
| Will dollar liquidity expand through June? | Kept | High value of resolution; clean data source. |
| Will funding conditions ease? | Merged | Correlated above 0.8 with the liquidity question. |
§ IIForecasting each question
Once the list is frozen, each question is sent to the forecasting agents independently. No agent sees another's output, and none sees the external probability. The Historian contributes base rates from comparable regimes. The Contrarian argues the other side. The forecasts are pooled as described in When Agents Disagree, producing a probability and a dispersion measure.
Separately, the Observer extracts the best available external probability for the same event: options-implied where a liquid market exists, swaps-implied for policy questions, survey-derived where nothing better is available. The source is published with every question, because a gap is only as meaningful as the reference it is measured against.
§ IIIFrom questions to a state
Ten independent probabilities are not, by themselves, a decision. The Allocator groups questions into the factors they load on, removes double-counting, and estimates the joint probability of the scenarios that would cause the most damage. The result is compressed into three published values.
- Regime: a two-part label. The first part describes the central case, the second its main vulnerability.
- Risk budget: a number between 0 and 1 expressing how much of the maximum permitted exposure the system is willing to carry.
- Conviction: Low, Moderate or High, derived from the size of admissible gaps and the dispersion between agents.
None of these is a recommendation. They are a description of the state of an autonomous system, published so that its reasoning can be followed and its record examined.
§ IVAccountability
Every outlook opens by scoring the previous one. Forecasts are frozen on issue and never revised. Each question, with our probability, the external probability and the outcome, enters the archive permanently. Over time, the archive — not this description — is the only honest answer to whether the process works.
The underlying sequence is the same one that runs continuously inside the system: observe, decompose, forecast, disagree, allocate, execute, learn. The outlook is simply the moment each month at which we stop and write it down.