Home » The Weekly Brief » No. 91
No. 91 · MethodWhat the analyst does for you — and the four places it fails
The terminal ships with a macro analyst. You state what you want watched, once. After that it checks on its own clock, acts, and writes down the number that caused it. This is what it does, what it refuses, and where its own arithmetic says it has nothing to offer.
What you are actually handing over
The work a macro trader does by hand is not the charting. It is the bookkeeping around the charting: re-running the same correlation every Monday to see whether it still holds, remembering which of five series was the strongest one last month, checking whether the thing that just crossed a level has ever crossed it before, and — the part everyone skips — going back to the study you ran in March to see whether it still says what you remember it saying.
That is what the analyst takes. You give it one sentence. In the terminal it is picked from four, in plain words:
- "Watch this market against inflation, liquidity and geo-risk, and tell me when the relationship changes"
- "Arm the conditions I care about and tell me the moment one of them is true"
- "Write me a briefing on a cadence, with what changed since the last one at the top"
- "Watch the relationships, arm what this screen implies, and brief me on a cadence"
Each one carries its own price, printed next to it before you choose. The last one reads: three jobs at once means three sources of interruption. It will speak more often than any single instruction would, and some of it will be about things you did not ask about.
What it does while you are elsewhere
On XAUUSD, that fourth instruction produces this in the standing desk inside the first two checks:
STANDING Standing instruction set: Watch the relationships, arm what this
screen implies, and brief me on a cadence
It runs every 6 seconds on this terminal's clock, on XAUUSD.
ARMED Armed CPI YoY above 3.2 %
half a point above where CPI is printing now. It reads CPI YoY
2.7 % now, so it is false, and it crossed 11 times in the
ten-year sample, last on 05 Nov 2024.
NOTHING Nothing worth telling you about
Checked 6 things: MENA escalation 88 · CPI YoY 2.7 % ·
INFLATION tone 0.9 · strongest link still CPI at +0.65 ·
1 armed condition unchanged · liquidity regime still Easing.
Nothing crossed a threshold I hold, so there is nothing to say.
On a macro desk that is the honest answer most days.Three things are worth pointing at. The armed condition arrives with a count — 11 crossings in ten years, 1.1 a year — so you know before you accept it whether it is a metronome or a bet on something new. The action is logged as something you can take back: one click undoes it, another stands the whole desk down, and anything it armed stays armed until you disarm it yourself. And the third entry is the one that matters most: it wrote a line to tell you that nothing happened.
The event study is the output, and its failures are the point
Ask the analyst what a CPI surprise does to gold and it does not answer with a direction. It defines the event as a rule over the series — a CPI print 0.2 points or more away from the one before it — finds every instance in the ten-year sample, and draws all of them.
STUDY CPI <GO>.The numbers it puts on the panel are +0.3 % median move over the twenty sessions after, 57 % went that way, −8.4 % furthest against, 61 observations. Under them, in the same type size as everything else: 26 of 61 went the other way — the spread runs from −8.4 % to +14.9 %. That is a description of 61 past instances on sample data. It is not a forecast, and 61 observations is far too few to trade a distribution from.
The five worst misses are listed by date, each one clickable, each one drawn in red on the fan. 18 Sep 2021, CPI 5.7 → 6.2 %, gold −8.4 % twenty sessions later. 18 Jun 2021, CPI 5.1 → 5.4 %, −7.0 %. A tool that shows only the clean examples is a horoscope.
The four places it fails
Ask it is this real? and it computes the scatter of every instance and the interval around the median, from the same array the fan is drawn from. On XAUUSD that verdict comes back negative four times out of five.
| Event class | n | Median +20d | Interval around the median | Verdict |
|---|---|---|---|---|
| CPI surprise | 61 | +0.3 % | −1.0 % … +1.6 % | contains zero — no direction |
| FOMC move | 29 | +1.9 % | +0.2 % … +3.6 % | excludes zero — the only one |
| Escalation crossing | 20 | −0.9 % | −3.4 % … +1.6 % | contains zero — no direction |
| Positioning extreme | 15 | +0.1 % | −2.9 % … +3.1 % | contains zero — no direction |
| Quarterly expiry | 40 | −0.1 % | −1.9 % … +1.7 % | contains zero — no direction |
XAUUSD, ten-year sample. Interval = median ± 1.96 standard errors of the median.
Four of the five classes the terminal ships with produce a median that sits inside its own error bar. When that happens the analyst leads with it: it contains zero. On this sample the class does not tell you a direction — 26 of the 61 went the other way, and that is what a median inside its own error bar looks like. Use the study to see the spread, not to pick a side.
This is not a flaw the desk is apologising for. It is the finding. A terminal that produced five confident directions from five event classes on one instrument would be fitting noise, and you would have no way of telling which of the five was the real one.
Where else it comes up empty
- Too few instances. Below four complete instances it draws nothing: any average from that would be a coincidence with a decimal point on it, so the terminal does not draw one. Ask anyway and it repeats the refusal rather than obliging.
- Inside the noise floor. Below |r| = 0.15 a correlation on this sample cannot be told apart from nothing, so it takes the series off the axis and says why. On gold's 90-session window the weakest of the five reads −0.12 — removed, not plotted.
- Conditions that never fired. Ask it to watch for the gold–escalation relationship changing sign and it arms
r to escalation below 0.00, then tells you that condition crossed 0 times in ten years — and that the relationship could decay from +0.56 to 0.05, dead in every practical sense, without ever crossing zero. There is no setting that catches both the sign change and the slow death. - Firing rate is not profit. Ask it to grade your armed conditions and it grades them on how often they cross, then adds: that is a count of crossings, not of profitable ones — no condition here has ever been tested for whether acting on it made money, and I am not going to imply it has.
What it will not do at all
Ask it how much to buy and the answer is the same every time: RegimeLens is a research terminal, not an execution or advice tool — I will not propose a trade size or a return. What I can do is show you the macro context: which series moves XAUUSD, whether it leads or lags, and what price did around the last policy or escalation event. It does not size positions, it does not place orders, and it does not name a return. Everything here is educational model output on illustrative data, not investment advice.
Reproduce it
Open the terminal, type START <GO> and answer the question. Or go straight to the numbers: STUDY CPI <GO> for the study above, REGIME GEO <GO> for the correlation split, STANDING <GO> for the desk. Every figure in this issue sits on one of those three panels, and the analyst reads them off the same arrays the canvas draws.