Forward Volatility: Units, Coordinates, Quote Ranges, and Event Variance

Extract an interval variance from two horizons while controlling units, day counts, surface coordinates, bid-ask ranges, event baselines, and tradability.

For educational purposes only; not investment advice. Investing may result in loss.

Direct answer

Forward volatility is an annualized volatility-rate summary inferred for a future interval. At one common valuation timestamp, let 0 < T1 < T2 be year fractions. Define annualized variance rate V(T) in decimal-squared per year and total variance W(T) = T x V(T) in dimensionless log-return-squared units. Then:

V_fwd(T1,T2) = [W(T2) - W(T1)] / (T2 - T1) and sigma_fwd = sqrt(V_fwd).

If the inputs are comparable annualized volatilities sigma(T), then W(T) = sigma(T)^2 x T. The subtraction is performed on total variance, not on volatility percentages. The output is model-, surface-, quote-, coordinate-, and convention-dependent; it is not a guarantee of realized volatility, a direction forecast, or automatically a tradable price.

Construct a comparable interval metric

Name the object being compared: a single-strike Black-Scholes IV, an ATM surface interpolation, a model-free implied-variance measure, a variance-swap strike, an index methodology, or an exchange-listed variance claim.
Freeze one valuation timestamp, exact expiration timestamps, time zone, calendar, and day-count and annualization convention; use year fractions rather than rounded month labels.
Capture synchronized bid, ask, size, forward, rates, dividends, borrow, settlement, and currency inputs for both horizons.
Fix a comparable coordinate such as k = ln(K/F), a documented Delta convention, or ATM-forward, plus the interpolation, smoothing, strike-truncation, and wing-extrapolation rules.
Convert each horizon into total variance W, compute the interval difference and annualized rate, and run nonnegativity, calendar-arbitrage, quote-envelope, and unit checks before taking a square root.
Distinguish the inferred scalar from a forecast, future smile, forward-start price, calendar spread, variance swap, VIX future, or variance future; stress events, jumps, baselines, surface rotations, and execution.
Preserve the data vintage and methodology, then reconcile any actual position’s fills, realized variance definition, settlement, margin, counterparty, fees, and tax separately from the diagnostic metric.

For a truly comparable additive expected-variance object, W(T2) - W(T1) is the interval total variance by construction. Treating two point IVs as that object is the model-dependent approximation. A variance-swap strike is quoted in variance units; a volatility-swap strike is not generally the square root of it because of convexity. VIX, VIX futures, and exchange-listed variance futures are also distinct claims and methodologies.

Two endpoints identify only an interval average variance rate. They do not identify the path within the interval, the future smile, future ATM IV, the joint distribution needed for a forward-start option, or realized volatility. Similarly, sqrt(E[variance]) is not generally E[sqrt(variance)].

Four worked examples

Exact day count: Under ACT/365, let T1 = 30/365, sigma1 = 40%, T2 = 90/365, and sigma2 = 30%. Then W1 = 0.0131506849315, W2 = 0.0221917808219, Delta W = 0.00904109589041, and Delta T = 0.164383561644. Thus V_fwd = 0.055000000000 and sigma_fwd = 23.4520787991%, not 40% - 30%.
Coordinate mismatch: Let T1 = 0.25 and T2 = 0.50. At matched ATM-forward coordinates, sigma1 = 20% and sigma2 = 25%, giving W1 = 0.010000, W2 = 0.031250, V_fwd = 0.085000, and sigma_fwd = 29.1547594742%. Mechanical use of one dollar strike whose IVs are 24% and 22% gives W1 = 0.014400, W2 = 0.024200, V_fwd = 0.039200, and sigma_fwd = 19.7989898732%. The difference is 9.3557696010 percentage points and reflects coordinate mismatch.
Quote envelope: With T1 = 0.25, front bid/ask volatility is 24% / 26%; with T2 = 0.50, back bid/ask is 25% / 27%. After mapping all four quotes into the same variance object, a conservative sell-side rate is [0.25^2 x 0.50 - 0.26^2 x 0.25] / 0.25 = 0.0574, or 23.9582971014%; the buy-side rate is [0.27^2 x 0.50 - 0.24^2 x 0.25] / 0.25 = 0.0882, or 29.6984848098%. Mid inputs 25% / 26% give 26.9629375254%. This is a diagnostic range, not a package-fill guarantee.
Event residual: A ten-day total window has sigma_total = 45%; the documented ordinary baseline for nine non-event days is sigma_base = 25%. Then W_total = 0.45^2 x 10/365 = 0.00554794520548, W_base = 0.25^2 x 9/365 = 0.00154109589041, and v_event = 0.00400684931507. The event-window log-return standard deviation is sqrt(v_event) = 6.3299678633%. Annualizing that one-day residual gives 120.9338662245%, but neither number is direction, probability, expected absolute return, or a guaranteed move.

Calculation and trading failure modes

The two inputs describe different variance, volatility, index, futures, or swap objects.
Surfaces are captured at different timestamps or after one market has moved.
Expiration times are reduced to dates and intraday fractions are lost.
Day count, annualization, calendar, decimal, volatility-point, or variance units are mixed.
Underlying, currency, exercise style, settlement, or observation convention differs.
The same dollar strike is mistaken for the same forward moneyness.
Forward, rates, dividends, borrow, or FX inputs are wrong or inconsistent.
Delta coordinates use different models, quote sides, or volatility inputs.
Quotes are stale, crossed, locked, zero-size, or not jointly executable.
Bid and ask inputs are combined with the wrong long or short orientation.
Sparse strikes, truncation, discrete integration, or bad quotes distort model-free variance.
Smile interpolation, smoothing, and wing extrapolation create hidden sensitivity.
Negative Delta W is silently floored rather than investigated.
ATM or single-strike IV is presented as model-free expected variance.
Variance-swap and volatility-swap strikes or their convexity adjustment are conflated.
VIX, VIX futures, variance futures, and a custom OTC claim are treated as interchangeable.
Event dates, ordinary baselines, overlapping catalysts, or window lengths are wrong.
Jumps, overnight returns, corporate actions, discrete monitoring, and variance risk premium are ignored.
One forward scalar is assumed to determine a future smile, joint law, or forward-start value.
Calendar-spread Greeks, assignment, execution, margin, OTC credit, settlement, fees, and tax are omitted.

Common misconceptions

“Subtract the two IVs.” Comparable total variances weighted by time are subtracted.
“Forward volatility is a guaranteed forecast of future realized volatility.” It is an implication of current inputs and assumptions.
“One scalar determines the future smile or forward-start price.” It identifies only an interval average under the chosen construction.
“A negative value means the market predicts negative volatility or guarantees arbitrage.” It is usually a data, coordinate, quote, interpolation, or methodology alarm.
“A calendar spread or VIX futures curve directly and purely trades this number.” Those positions have different claims, surfaces, Greeks, execution, and lifecycle risks.