Risk-Neutral Vs Real-World Density

Pull a probability distribution out of option prices — it's a well-known trick, using the Breeden-Litzenberger Formula — and you'll get a curve that looks like a forecast: a most-likely price, fat or thin tails, skew one way or the other. It is tempting to read that curve as "what the market thinks will happen." That is a mistake, and a costly one if you trade on it, because the curve you extracted is not a forecast at all.

An insurance company's price list isn't a weather forecast

An insurer's flood-policy pricing reflects how likely a flood is and how much it would hurt to be flooded without coverage — a flood in a place with no drainage costs more to insure than an equally probable flood somewhere the water just runs off. The insurer's price list is not a pure weather forecast; it's a forecast bent by how painful the bad outcome is. Option prices work the same way. They embed both the market's actual belief about where the stock might go (the real-world or physical density, written $P$) and a markup for how painful it would be if the bad outcome happened (risk aversion). What Breeden-Litzenberger extracts from option prices is the blend of the two — the risk-neutral density $Q$ — not the pure forecast $P$.

The formula

$$\frac{dQ}{dP} = \frac{e^{-\gamma \, U'(S_T)}}{\mathbb{E}^P[e^{-\gamma\, U'(S_T)}]}$$

In plain English: $dQ/dP$ is the "exchange rate" between the two densities — it says how much extra weight $Q$ puts on an outcome compared to $P$. That exchange rate is driven by risk aversion, $\gamma$, and the marginal utility $U'(S_T)$ of wealth in that outcome. When the stock is very low (a painful, low-utility outcome), $U'$ is high, so $Q$ puts more weight on that outcome than $P$ actually believes is likely — the market prices crash protection as if crashes were more probable than they really are, because investors hate crashes disproportionately to their true odds. A simpler, model-free intuition: $Q$ equals $P$ multiplied by a "fear factor" that is large in bad states and small in good ones.

Worked example 1: comparing implied crash odds to realized ones

Suppose one-year S&P 500 options imply a risk-neutral probability of a 20% market decline of about 8% (a number you could extract via Breeden-Litzenberger from the observed put prices at that strike). Historically, looking at actual market data over many decades, 20% one-year declines have occurred roughly 3–4% of the time — call it 3.5%. The ratio $8\% / 3.5\% \approx 2.3$ is a rough read on the variance risk premium: option markets price crash protection at more than twice the frequency crashes have actually occurred, because sellers of that protection demand compensation for bearing a painful, correlated risk, not because they genuinely believe crashes are 2.3 times more common than history suggests.

Worked example 2: a two-state toy model made concrete

A stock at $100 goes to $130 (good state) or $80 (bad state) in one year; the real-world probability of the bad state is $P(\text{bad}) = 30\%$. The risk-free rate is 0%, and suppose the risk-neutral probability implied by option prices is $Q(\text{bad}) = 45\%$ (options are pricing the bad state as more likely than it actually is, i.e. paying a premium for downside protection). The ratio $dQ/dP$ in the bad state is $0.45/0.30 = 1.5$ — the market's "fear multiplier" for that outcome is 1.5×. In the good state it must be $dQ/dP = 0.55/0.70 = 0.786$, below 1, since the two densities must both sum to 1 and $Q$ over-weights bad, it necessarily under-weights good. A put struck at $100 pays $20 in the bad state and $0 otherwise, so its price is $Q(\text{bad}) \times 20 = 0.45 \times 20 = 9.00$, i.e. $9.00 — noticeably above the $6.00 you'd get by naively using the real-world probability, $0.30 \times 20$. That $3.00 gap is the price of risk aversion, paid by anyone who wants insurance against the bad state.

Picture two overlapping bell curves on this explorer: the real-world density centered where analysts genuinely expect the stock, and a risk-neutral density shifted and fattened on the left tail — same shape family, different weight on the bad outcomes.

What this means in practice

Treating an option-implied density as an actual forecast is a common and expensive error: it systematically overstates crash probabilities (great if you're pricing insurance, terrible if you're trying to predict the future). Volatility arbitrage desks exploit exactly this gap — selling the persistently overpriced downside protection embedded in $Q$ while trying to stay hedged against the real-world path the stock actually takes, which is the economic logic behind the variance risk premium harvested by VIX vs Realised Volatility Spread Trades.