Volatility Trading Explained: Trading Vol as an Asset Class

Volatility trading isolates the magnitude of price movement as a tradable quantity, independent of direction. For a quant the core object is the spread between implied volatility (IV) — the market's risk-neutral forecast embedded in option prices — and the realized volatility (RV) that subsequently occurs. Systematic vol trading is the business of forecasting RV, comparing it to IV, and harvesting the structurally positive but negatively-skewed variance risk premium (VRP) while surviving its tail. This guide is about the estimators, the hedging error, and the failure modes — not the option-strategy zoo.

The central asymmetry: on average IV > RV (you are paid to sell insurance), but the payoff to selling is concave and the rare loss is enormous. Every design decision flows from that one fact.

IV, RV, and the right estimators

RV is not a single number; your choice of estimator changes the trade. Close-to- close understates intraday movement; range-based estimators (Parkinson, Garman-Klass, Yang-Zhang) are far more efficient for the same sample. For forecasting RV — the thing you actually trade against IV — a GARCH model captures volatility clustering and mean reversion that a trailing window misses.

import numpy as np
import pandas as pd

def yang_zhang_rv(df, window=20, periods=252):
    """Yang-Zhang realized volatility: efficient, handles overnight gaps.
    df columns: open, high, low, close."""
    o, h, l, c = np.log(df.open), np.log(df.high), np.log(df.low), np.log(df.close)
    oc = o - c.shift(1)                          # overnight (close-to-open)
    co = c - o                                   # intraday (open-to-close)
    rs = (h - c) * (h - o) + (l - c) * (l - o)   # Rogers-Satchell term
    k = 0.34 / (1.34 + (window + 1) / (window - 1))
    vo = oc.rolling(window).var()
    vc = co.rolling(window).var()
    vrs = rs.rolling(window).mean()
    return np.sqrt((vo + k * vc + (1 - k) * vrs) * periods)

The tradable signal is the VRP: IV - forecast_RV, often as a ratio IV / RV. A VRP that is high and a forecast that RV will stay low is the short setup; a collapsing VRP with a forecast of rising RV is the long setup. The measuring volatility piece covers estimator trade-offs in depth.

The variance risk premium is real but conditional

Selling options is profitable on average because demand for crash protection is persistent and price-inelastic — institutions overpay for downside hedges the way people overpay for insurance. But the VRP is not constant: it is compressed in calm markets (little premium left to harvest, yet everyone is short) and explodes after a shock (rich premium, but you are already wounded). The information ratio of naive short-vol is flattered by samples that exclude a crash; the strategy's true distribution is many small gains and occasional ruinous losses.

Side	Greeks	Payoff shape	VRP
Long vol (buy straddle)	long vega, long gamma, short theta	small steady losses, rare big wins	pays the premium
Short vol (sell premium)	short vega/gamma, long theta	small steady gains, rare big losses	earns the premium
Gamma scalp (long gamma, delta-hedged)	long gamma, short theta	monetizes RV > IV via re-hedging	net of hedging cost

Gamma scalping: monetizing RV - IV directly

The cleanest expression of the RV-vs-IV view is delta-hedged long gamma. You own options and re-hedge delta as the underlying moves; each re-hedge locks in a small gain proportional to the squared move. The theoretical PnL of a continuously delta-hedged option over a step is approximately:

dPnL ≈ 0.5 * gamma * S^2 * (realized_variance_step - implied_variance_step)

You profit when realized variance exceeds implied by enough to cover the theta you pay and — critically — the transaction costs of hedging. The hedging frequency is a real parameter with a real trade-off: hedge too often and costs eat the edge; hedge too rarely and you carry path risk and tracking error to the continuous-hedge ideal. This discrete-hedging error is a first-order source of slippage between the textbook formula and live PnL. The Greeks themselves are in options Greeks; the underlying diffusion assumptions in Black-Scholes.

Where the assumptions break live

The neat RV/IV framework rests on assumptions that fail in exactly the regimes that matter:

Jumps, not diffusion. Gamma-scalping PnL assumes continuous paths;

overnight gaps and jumps cannot be hedged and deliver losses the model says are impossible. Short gamma is short jump risk.

Skew and the wrong volatility. A single "IV" is a fiction — the surface has

skew (index puts richer than calls, the "smirk") and term structure. You are never short the vol; you are short a specific point on a surface that reshapes under stress.

Vega is not stationary. As markets move, your vega and gamma drift, so a

"vega-neutral" book at inception is directional in vol within a day.

Correlation goes to one. In a crash, "diversified" short-vol positions all

lose together; cross-sectional diversification evaporates when you rely on it.

The VIX complex and its traps

You cannot trade the VIX index — only futures, options, and ETPs on it, which roll along the VIX futures term structure.

Term structure: contango (longer-dated IV higher) is the calm norm;

backwardation (near-dated higher) signals stress.

Skew: persistent put richness from crash-hedging demand.

The structural trap: long-VIX ETPs roll up the contango curve and decay relentlessly in calm markets, while short-VIX ETPs harvest that roll until a spike annihilates them. February 2018 "Volmageddon" liquidated several inverse-VIX products in a single session when VIX more than doubled — a direct consequence of short-gamma exposure embedded in a levered, daily-rebalanced wrapper. Crypto has DVOL (BTC/ETH implied-vol indices) and listed options with the same skew and term-structure dynamics. Calendar spreads trade the term structure; risk reversals trade the skew — both are views on the shape of the surface, not its level.

Sizing short gamma: the only thing that matters

Short vol is short gamma, so losses accelerate as the market moves against you — the position gets shorter exactly when it is losing. Variance-based sizing is therefore actively dangerous; size off the tail.

def short_vol_stress(notional_vega, iv_now, iv_shock_mult=2.0, gamma_pnl_per_pt=None):
    """Crude stress: vega loss from an IV doubling, before convex gamma losses."""
    iv_jump = iv_now * (iv_shock_mult - 1)
    vega_loss = notional_vega * iv_jump          # linear vega term
    return vega_loss                             # real loss is worse: gamma is convex

Defined-risk structures (iron condors, vertical/wing spreads) cap the tail by buying farther-out options — paying away some premium for a known maximum loss. For a professional, undefined-risk short premium is rarely justified: the left-tail loss is unbounded and the CVaR is what kills the book. Frame sizing with max drawdown and explicit position-sizing on the worst plausible day, hold tail hedges, and keep dry powder for the rare windows when IV is genuinely cheap.

Honest edge assessment

The VRP is one of the more durable premia, but it is crowded and its Sharpe is sample-dependent — measure it across a window that includes a crash or the number is fantasy. Validate with deflated Sharpe given how many strike/expiry/hedge-frequency combinations you can search, and remember the strategy's negative skew means a high backtest Sharpe is partly "unpaid claims." The edge is real on average and over-rewarded in calm; the discipline is not being maximally short into the regime where the premium is about to be paid back with interest.

Conclusion

Volatility trading monetizes the IV-minus-RV spread and the structural variance risk premium, with gamma scalping the purest long-side expression and premium selling the short side. The premium is real but conditional and negatively skewed; the math that matters is RV estimation, discrete-hedging error, surface skew and term structure, and — above all — sizing short gamma for the jump-driven tail rather than the diffusion-driven average. Done with defined risk, honest crash-inclusive backtests, and active Greek management, it is a genuinely orthogonal return stream; done with naked premium and variance-based sizing, it is a fast, well-documented way to blow up.