What Volatility Knows: The Market's Forecast of Its Own Uncertainty, and How to Read It

Volatility is the only thing a market forecasts about itself. Prices carry expectations about earnings, inflation, and policy, but those expectations are about the world; the market's expectation of its own future movement is written down directly, in the prices of options, and can be checked against what subsequently happens. That makes volatility unusual among market quantities. It is observable after the fact, it is forecast before the fact, and the gap between the two is a price that someone pays and someone collects.

This piece is about what is embedded in that forecast and its neighbors: the realized volatility computed from past returns, the implied volatility read from options, the way implied volatility varies with tenor and with strike, and the way volatility in one market moves with volatility in others. Each of these knows something different. None of them knows everything, and the useful information is usually in where they disagree. Throughout, the figures are stylized simulations built from simple models stated in the notes; they are meant to make the mechanisms visible, not to describe any particular market or period.

Two Numbers and the Premium Between Them

Realized volatility is a fact about the past: take a window of returns, compute their dispersion, annualize. Implied volatility is a price about the future: the volatility that, fed into an option pricing model, reproduces the price at which an option currently trades. The two are often compared as if they were forecast and outcome, and to a first approximation they are. But an option is insurance, and insurance is not sold at its expected cost. The buyer pays for the removal of a risk and the seller charges for bearing it, so implied volatility contains a forecast and a premium, and the two are not separately labeled.

Figure 1 shows what this looks like in a stylized model. Each point is a simulated month; its horizontal position is the implied volatility at the start of the month and its vertical position is the realized volatility over the month that followed. The dashed line is the 45-degree line along which forecast equals outcome. Most points sit below it: implied volatility exceeded what was subsequently realized, and the seller of insurance collected a premium. A minority sit above it, some well above, and those are the months in which the premium was not enough.

The shape of the cloud is the information. The premium is not a constant offset; in the model it scales with the level, so that the gap between forecast and outcome widens in absolute terms when volatility is high. The one-sided gap is the signature of a risk premium rather than a forecast error: an unbiased forecast would scatter symmetrically about the line, however wide that scatter grew with the level. A systematic investor who reads implied volatility as a pure forecast will be persistently surprised in one direction; one who reads it as a pure premium will be occasionally surprised, badly, in the other. The useful reading is that implied volatility is a forecast plus a price for being wrong, and that the price is set by how badly the marginal buyer wants not to be wrong.^[1]

The Term Structure Is a Forecast of Forecasts

Options trade at many expiries, and each expiry has its own implied volatility. Read across tenors, the result is a term structure, and it says something a single tenor cannot: not only how much movement the market expects, but how that expectation is distributed over time. In calm conditions the term structure typically slopes upward. Near-term volatility is low because nothing is scheduled to happen; longer-dated volatility is higher because more can happen, and because sellers of long-dated insurance charge for the uncertainty of their own estimate. Under stress the curve inverts: near-term volatility rises sharply, longer-dated volatility rises less, and the shape encodes an expectation that whatever is happening will not continue at this intensity for long.

Figure 2 draws both states in a stylized model in which implied volatility at each tenor decays exponentially from a short-end level toward a long-run level. The calm curve rises from the short end toward its anchor; the stressed curve falls toward a higher anchor. The horizontal line is a stylized long-run average of realized volatility. The stressed curve sits above it at every tenor. The calm curve starts below it and crosses it between three and six months: in calm conditions the market's forecast for the next few weeks is below the unconditional average, and the premium of Figure 1 is measured against that conditional forecast, not against the long-run line. Beyond six months the calm curve, too, sits above the line, which is the premium again, now visible as a function of horizon.

Two readings follow. The first is that the slope of the term structure is a statement about persistence. An inverted curve says the market expects mean reversion; the speed of the decay in the figure, set by the parameter in the note, is the market's estimate of how quickly. Whether that estimate is right is a separate question, and in a stylized sense it is the same question as whether a point in Figure 1 lands above or below the line. The second reading is that the long end moves too. Stress does not only lift the front of the curve; it raises the level the curve decays toward, because a period of high volatility updates the market's view of what normal is. The long end is the slow-moving memory of the option market, and it is often more informative about the regime than the front end is about the week.

Skew Is the Price of Direction

Implied volatility also varies across strikes at a single tenor, and that variation, the skew or smile, carries a different kind of information. A model in which returns are symmetric and volatility is constant would produce a flat line across strikes. Observed markets do not. In equity indices, out-of-the-money puts trade at higher implied volatility than out-of-the-money calls: the market pays more to insure against a fall than against a rise of the same size, partly because falls tend to be faster and to coincide with rising volatility, and partly because the natural holders of equities are long and want protection on one side only. In some commodity markets the tilt runs the other way; consumers of the commodity fear a spike, and calls carry the premium. In many currency pairs the smile is close to symmetric, because there is no natural long side to the market.

Figure 3 draws three stylized smiles from a single quadratic model in log-moneyness with different parameters: an equity index with put skew, an energy contract with call skew, and a currency pair with a near-symmetric smile. The vertical line marks the at-the-money strike.

The slope of the skew measures asymmetry in the market's expectation and asymmetry in who needs insurance. The curvature measures how much weight the market puts on large moves in either direction relative to small ones. Both change over time, and their changes are informative in a way the at-the-money level alone is not. A steepening put skew with an unchanged at-the-money level says that the demand for protection has shifted toward the tail without the center of the distribution moving. That is a different state of the world from a rise in the level with an unchanged skew, and a process that watches only the at-the-money number treats the two as identical.^[2]

Volatility Travels

The measures discussed so far are properties of one market. The last is a property of the relationship between markets. Volatility co-moves across asset classes far more strongly than returns do. Returns can move in opposite directions, and a good deal of portfolio construction depends on that; volatility, in nearly every market, tends to rise together. A shock to risk appetite raises the expected movement of equities, credit, currencies, and commodities at once, whatever it does to their levels. This is why the correlation of changes in volatility across markets is almost always positive, and why it rises in stress, while return correlations may go in any direction.

Figure 4 shows a stylized correlation matrix of volatility changes across seven markets, generated from a two-factor model: a global risk-appetite factor that every market loads on with the same sign, and a policy-and-rates factor that loads most heavily on interest rate volatility and on the markets that trade off it. The matrix is entirely positive. Its structure is in the magnitudes: the equity, credit, and emerging-market block moves nearly as one; rates and gold form a second, looser group tied together by the policy factor; and crude oil sits somewhat apart, driven more by its own supply dynamics than by either factor.

What the matrix knows is where a shock is coming from. When volatility rises in one block and not the other, the source is local; when it rises everywhere, the source is a common factor, and the appropriate response is different. A local volatility shock is information about one market and can be weighed against the others. A global one is information about the risk budget itself, because every position in a portfolio, whatever its sign, has just become larger in risk terms. The distinction is not visible in any single market's implied volatility. It is visible only in the co-movement.^[3]

What We Do With It

Oak St. treats volatility as data with a structure, not as a single number to be forecast. Each measure enters the research process as the answer to a different question, and the questions are kept separate. Realized volatility tells us what happened and at what frequency; implied volatility tells us what the market will pay to be insured, which is a forecast and a premium together; the term structure tells us how long the market expects the present state to last; the skew tells us which direction it fears; and cross-asset co-movement tells us whether a shock is local or shared. Figure 5 sets these out side by side, with the limits of each, because the limits are where the mistakes happen.

Three practices follow. The first is to model the premium separately from the forecast, so that a comparison of implied and realized volatility is a statement about the price of insurance rather than an accusation that the market cannot forecast. The second is to read shapes before levels: the slope of the term structure, the slope and curvature of the skew, and the block structure of the co-movement matrix each change in ways the headline level does not reveal, and the changes are where the information tends to be. The third is to treat a shared rise in volatility as a change in the size of every position rather than as a signal about any one of them, and to size accordingly, before the change in size becomes a change in loss.

Volatility knows a great deal. It knows what happened, what the market will pay to avoid a repeat, how long it expects the present to last, which direction it fears, and whether the fear is local or shared. It does not know the future, and the most reliable thing it says about the future is the price at which someone is willing to insure against it. That price is worth reading carefully, in all of its dimensions, precisely because it is a price and not a prophecy.