In deeply cyclical sectors the P/E ratio inverts: it peaks at the bottom and troughs at the top. Understanding why turns the multiple from a misleading signal into a useful one.
Most people read a P/E ratio as though only the numerator moves. Price goes up, the stock gets expensive; price goes down, it gets cheap. For a stable business that is a workable approximation, because earnings grind along a fairly smooth path and the multiple mostly reflects sentiment about price.
Semiconductors are not a stable business. Aggregate sector earnings can double and then halve inside a couple of years, driven by a capacity cycle that overshoots in both directions. When the denominator moves that hard, it stops being a passenger and starts driving.
The result is that the semiconductor P/E does not measure what most readers think it measures.
Follow a full cycle. Demand is strong, fabs are running flat out, margins are fat, and earnings are at a peak. Investors, aware that this cannot last, decline to pay a high multiple on peak profits. Price is high, earnings are higher, and the P/E is low.
Now the cycle rolls over. Orders slow, inventories build, pricing cracks. Earnings fall much faster than the share price does, because investors are already looking across the valley to the recovery. Price is down, earnings are down far more, and the P/E is high.
So the multiple is lowest near the top and highest near the bottom. An investor screening for "cheap semiconductors on a P/E basis" is, mechanically, screening for peak-cycle exposure. This is not a subtlety or an edge case; it is the ordinary behaviour of the ratio in this sector.
The inversion is the mild version of the problem. The severe version arrives when aggregate expected earnings approach zero, which for this sector is not hypothetical — it happened in the depths of the financial crisis.
As the denominator shrinks toward zero, the ratio does not become large in an orderly way. It accelerates toward infinity. A sector multiple can print in the hundreds, then in the thousands, over a single quarter in which the actual news barely changed. And when expected earnings cross zero into negative territory, the P/E does not become a big number. It ceases to exist.
This is worth sitting with, because it has practical consequences beyond one ugly chart. Any statistic computed across such a period is contaminated: a five-year average, a percentile rank, a z-score. A single reading in the thousands will dominate an average that includes it, and the model consuming that average will confidently report nonsense.
Invert the ratio. Expected earnings divided by price gives an earnings yield, and it is well behaved everywhere the P/E is not. As earnings fall toward zero the yield simply approaches zero. As they cross into losses it goes negative, which is both meaningful and finite. Nothing explodes; nothing becomes undefined.
This is why our valuation pages carry the earnings yield alongside the multiple rather than treating it as a footnote. For three of the four indices we track it is a convenience. For semiconductors it is the only representation that survives a full cycle.
The second habit worth building is to separate the two moving parts explicitly. Watching expected earnings as its own series, next to price, tells you immediately whether a falling multiple is the market marking something down or analysts marking something up. The ratio alone cannot distinguish these, and they call for opposite responses.
The practical test of any framework is what it says about today. As of August 2026 the semiconductor sector trades on a blended forward multiple in the low twenties — neither remarkable nor obviously stretched by the standards of the last two decades.
The interesting part is not the level but its stability. Over the year into that reading, the sector index rose sharply and expected earnings rose by a comparable amount. The multiple went roughly nowhere.
That is the signature of an earnings-driven advance rather than a valuation-driven one, and it is a materially different situation from a rally of the same magnitude delivered by multiple expansion alone. Which of the two you are in cannot be read off the P/E, because in both cases the P/E does the same thing: very little. It can only be read by looking underneath it.
Deep Dive
The full methodology behind the June 2026 rebuild: the net-liquidity spine, the anchor-free flow estimator, the noisy-OR funding override, and why the headline became an absolute 0-100 reading instead of a rolling percentile.
Deep Dive
The reserve-demand curve is convex, so a drain into a thin cash cushion is not the same event as a drain into an ample one. How DLI 2.0 prices that asymmetry without falling into the reserve-level trap that a percentile weight fell into.
Deep Dive
Bitcoin's correlation with Fed liquidity is one of the strongest macro relationships in modern markets. Here's the mechanism in plain language, the historical pattern from 2018, 2020, 2022, and the single indicator that has flipped first ahead of every BTC bottom.