Mid cycle and late cycle produce almost identical headlines. Growth is positive in both. Unemployment is low in both. Earnings are usually rising in both. The market is usually up in both. That similarity is not a failure of anybody's data, it is the actual nature of the thing, and it is why the confident cycle-positioning takes you read are almost always written about a period that has already ended.
What follows is my attempt to be strict about it. Which readings on the Macroeconomic Risk Scorecard genuinely carry information about where you are in the cycle while you are still in it, and which ones look like they do and only resolve afterwards.
The tile gives you a phase, not a position inside it
Open the regime tab and you get one word describing the business cycle phase. Right now it says SLOWDOWN, sitting beside a combined M7 score of 29 out of 100 labelled LOW risk, a health grade of C at 59, and a counter for how many of the seven recession models are at or above 60.
Note what that is and is not. It is a classification of the current state. It is not a percentage through the phase, and the vocabulary you brought with you, mid cycle and late cycle, is your framing rather than a label I can see the module printing. So the honest starting position is that this is a question you answer from the underlying readings, not one you read off a tile. Anyone telling you a dashboard will tell you it is late cycle is describing a product I have not seen.
The three readings that discriminate while it is happening
Out of the fifty plus indicators feeding this thing, three families change character between the middle and the end of a cycle in a way you can see in close to real time.
Model disagreement. This is the best of the three and the one the module makes easiest. The count of models at or above 60 can rise while the combined score barely moves, because an average hides a split. Zero of seven elevated at a composite of 29 is a genuinely quiet reading. Three of seven elevated at a composite of 29 is a market where a meaningful minority of the evidence has already turned and is being averaged away. The second one is what the end of a cycle tends to look like from the inside. Not everything breaking, a few things breaking while the aggregate stays polite.
Credit against growth. The scorecard tracks credit default swap spreads and high-yield spreads on one side and growth and labour data on the other. Those two live on different clocks. Credit is priced continuously by people whose job is to be repaid, growth data arrives monthly and gets revised for years. When credit is deteriorating while growth prints are still fine, you have the one combination that is genuinely more common late than mid. It is not a prediction and it does not tell you when. It is a divergence, and divergences are information.
Policy direction rather than policy level. The module carries a Policy tab and tracks Federal Reserve settings, Treasury yields and the yield curve. The level of rates tells you very little about where in the cycle you sit, because rates are high in plenty of healthy periods. The direction and the reason are different. A central bank that has stopped tightening and is being asked publicly whether it will start cutting is in a different part of the argument from one that is still raising into an expansion.
The four readings everybody quotes that resolve too late
Now the discipline half, because knowing which of your inputs are useless in real time is worth more than adding another one.
- Unemployment. Low right up until it is not, and the initial prints revise. By the time the labour data is unambiguous the question you were asking has been answered by the market.
- Any peak. Peak margins, peak earnings growth, peak sentiment. A peak is defined by what came after it. You cannot observe one in the present, only in a chart with the right-hand side already filled in.
- Valuation. Expensive markets get more expensive for long stretches and expensive is not a phase indicator. It tells you about future long-run returns and almost nothing about the next twelve months.
- Official cycle dating. Announced with a substantial lag by construction, because the committees involved wait for revised data before they commit. Useful for history, useless for a decision you are making on a Sunday evening.
The uncomfortable summary is that the four inputs most people build their cycle view from are precisely the four that only work with hindsight. That is not an accident. They are the ones that produce clean charts afterwards, which is why they end up in the articles.
The decision this actually supports on a real account
Say twenty thousand dollars, sixty five percent broad equities, fifteen percent crypto, twenty percent cash, and you have decided the discriminating readings are leaning late rather than mid. What changes.
Not the allocation, mostly. The single most expensive mistake available here is going defensive on a late-cycle call and being two years early, because a market can keep rising for a long time after the discriminating readings start leaning, and sitting in cash through that stretch does more damage to a twenty thousand dollar account than the drawdown you were trying to avoid. Being early is not a smaller version of being right. It is its own loss.
What should change is cheaper and more specific.
- Stop adding leverage. If you have margin or perp exposure, do not increase the notional. Costs nothing, reversible on any day you change your mind.
- Rebalance back to your written weights instead of letting the winner ride. If crypto has drifted from fifteen percent to twenty six percent, that is roughly twenty two hundred dollars of exposure you never chose to take, and a late-cycle read is a good reason to take the decision back.
- Get the cash buffer to three to six months of actual spending, held outside the trading account. This is the holding that turns a drawdown into an inconvenience instead of a forced sale, and its value does not depend on your cycle call being right.
- Name the one position that would hurt most in a broad thirty percent decline and size it so that outcome is survivable. Usually the most growth-sensitive thing you own, and usually you already know which one it is.
Every item on that list costs you very little if the cycle turns out to have years left. That asymmetry is the whole reason to act on a signal this noisy. If your response to a late-cycle read is expensive when you are wrong, you have made the response too big for the quality of the evidence behind it.
What to do when the readings contradict each other
They will, and the contradiction is more informative than either reading alone. Model disagreement rising while credit spreads stay tight is a case where the models are picking up growth and labour softness that the people lending money have not repriced. Credit widening while the model count stays at zero is the reverse, and credit is generally the faster of the two, so that combination deserves more attention than its apparent smallness suggests.
What you should not do is average them into one number and act on the average, which is exactly what the composite already does for you. The reason to look past the 29 is that its inputs disagree, and collapsing the disagreement back into a single verdict throws away the only part of the reading you could not have got from the headline. Hold both, act on the cheap items above, and accept that the confident version of this call is only available to people writing about it afterwards.