Market-Based Variance Overview

• Market-based variance is defined as variance extracted directly from market observables rather than historical covariances.
• It employs methodologies like risk-neutral term structure inference, modified option payoff formulas, and aggregation of trade data into synthetic portfolios.
• This approach enhances model calibration, hedging accuracy, and portfolio risk estimation while addressing traditional misconceptions in variance measurement.

Searching arXiv for recent and foundational papers on “market-based variance” and closely related uses. arxiv_search query="market-based variance OR market implied variance term structure option returns variance portfolio trade volumes variance swap" max_results=10 sort_by="relevance" Market-based variance denotes variance quantities that are anchored in market observables rather than in a purely historical sample covariance. Across the literature, the label is used for several closely related objects: the risk-neutral term structure of expected future variance implied by traded options or variance swaps; the variance of discounted option payoffs computed under market-consistent inputs such as implied volatility; and portfolio or market variance reconstructed directly from actual trade values and trade volumes by treating a portfolio, or even the entire market, as a synthetic traded security (Yoo, 9 Sep 2025, Ben-Meir et al., 2012, Olkhov, 15 Oct 2025). A common theme is that the relevant variance is inferred from prices, option surfaces, or transaction flows that are already embedded in the market.

1. Conceptual scope and definitions

A central usage defines market-based variance as the term structure of expected future variance under the risk-neutral measure. In a jump-diffusion setting with stochastic variance $V_t$, realized variance over $[0,\tau]$ is the quadratic variation of log prices, and the annualized expected variance is

$$VS(\tau;\Theta) = \frac{1}{\tau}\,\mathbb{E}_0^{\mathbb{Q}_\Theta}[QV(\tau)].$$

The corresponding market-based quantity is the curve $\tau \mapsto VS^{mkt}(\tau)$ when variance swaps trade, or the curve $\tau \mapsto VIX^2(\tau)$ when it is inferred from options (Yoo, 9 Sep 2025).

A second usage treats variance as a contract-level risk measure. In the Black–Scholes framework, the discounted payoff of an option is a random variable, and its variance, higher moments, and the probability of expiring worthless can be computed explicitly for European vanillas and some barrier options, or via modified Black–Scholes PDEs for American puts (Ben-Meir et al., 2012). In this sense, market-based variance is the variance implied by the same market inputs that support pricing.

A third usage reconstructs variance from actual trade sequences. Here the investor does not trade after portfolio formation, but observes current trades in the constituent securities. Those trades are normalized and aggregated into a synthetic trade series for the portfolio, so that the portfolio has the same formal description as a single traded security. Variance then depends not only on price changes but also on the randomness of trade volumes and on value–volume covariance (Olkhov, 10 Apr 2025, Olkhov, 15 Oct 2025).

2. Option-implied variance term structures and variance swaps

In the options literature, the market-based variance term structure is closely tied to variance swaps, log contracts, and VIX-style replication. For continuous price paths, the canonical model-free variance swap formula writes the market-implied annualized variance as

$$VS^{mkt}(\tau) = \frac{2}{\tau}\left[\int_0^{F_\tau}\frac{P_0(K,\tau)}{K^2}\,dK + \int_{F_\tau}^{\infty}\frac{C_0(K,\tau)}{K^2}\,dK\right].$$

With jumps, the variance swap rate is no longer literally equal to $VIX^2$; the gap depends on jump asymmetry, and negative jump skewness implies $VS^{mkt}(\tau) > VIX^2(\tau)$ (Yoo, 9 Sep 2025). This distinction is operationally important in calibration: joint calibration of a parametric model to both the implied-volatility surface and the variance term structure materially improves the fit to observed VIX-like curves, and the practical recommendation is $\alpha \in [0.75,0.9]$ in the augmented objective that trades off smile fit against variance-curve fit (Yoo, 9 Sep 2025).

A model-independent strand derives variance-swap information directly from finitely many option prices and no-arbitrage. For weighted variance swaps, the payoff can be transformed into a convex European payoff $\lambda_T(S_T)$ plus dynamic trading, using a pathwise Itô relation with $[0,\tau]$0. The fair weighted variance swap rate $[0,\tau]$1 is then linked to the convex-payoff price $[0,\tau]$2 by

$[0,\tau]$3

and robust lower and upper no-arbitrage bounds follow from semi-infinite linear programming over measures consistent with observed put prices (Davis et al., 2010). Empirically, quoted variance swaps were found to lie surprisingly close to the model-free lower bounds (Davis et al., 2010).

A structurally richer interpretation appears in time-changed defaultable models. There the stock is modeled as a killed diffusion subordinated by a Lévy clock, and the Laplace exponent $[0,\tau]$4 of the subordinator is recovered from the option surface through a spectral inversion formula. This makes the time change, and therefore the variance dynamics and the variance-swap term structure, largely market-implied rather than exogenously fixed (Lorig et al., 2012). Empirically, SPX studies based on VIX futures likewise extract instantaneous variance curves

$[0,\tau]$5

and study their factor structure, daily changes, and joint dynamics with spot returns (Ségonne, 2015).

3. Contract-level option variance and market-consistent risk

Within Black–Scholes-type pricing, the usual price is only the first moment of the discounted payoff. The variance of a European call or put is obtained from explicit formulas for $[0,\tau]$6 and $[0,\tau]$7, with

$[0,\tau]$8

The second moment satisfies a modified Black–Scholes PDE,

$[0,\tau]$9

while the probability of expiring worthless satisfies the same spatial operator with zero discount term (Ben-Meir et al., 2012).

The same machinery extends to a European down-and-out put, where closed forms are obtained for the expected payoff, second moment, and PEW, and to an American put, where the price, second moment, and PEW satisfy separate PDEs on the continuation region with the same free boundary $$VS(\tau;\Theta) = \frac{1}{\tau}\,\mathbb{E}_0^{\mathbb{Q}_\Theta}[QV(\tau)].$$0. The American second moment again carries the reaction term $$VS(\tau;\Theta) = \frac{1}{\tau}\,\mathbb{E}_0^{\mathbb{Q}_\Theta}[QV(\tau)].$$1, and PEW is governed by a homogeneous backward PDE with absorbing boundary at the exercise frontier (Ben-Meir et al., 2012).

These constructions yield a buyer-side risk interpretation. Variance measures the dispersion of discounted payoff around its mean, while PEW measures the risk of complete loss. The paper also emphasizes the ratio $$VS(\tau;\Theta) = \frac{1}{\tau}\,\mathbb{E}_0^{\mathbb{Q}_\Theta}[QV(\tau)].$$2 as “risk per unit expected return” and shows that plugging implied volatility into the closed forms or PDEs produces market-implied variance and PEW for each contract (Ben-Meir et al., 2012). A further implication is that if option prices are adjusted by

$$VS(\tau;\Theta) = \frac{1}{\tau}\,\mathbb{E}_0^{\mathbb{Q}_\Theta}[QV(\tau)].$$3

then $$VS(\tau;\Theta) = \frac{1}{\tau}\,\mathbb{E}_0^{\mathbb{Q}_\Theta}[QV(\tau)].$$4 generates a volatility smile and $$VS(\tau;\Theta) = \frac{1}{\tau}\,\mathbb{E}_0^{\mathbb{Q}_\Theta}[QV(\tau)].$$5 a volatility frown when the adjusted prices are re-expressed as Black–Scholes implied volatilities (Ben-Meir et al., 2012).

4. Trade-based variance of portfolios, market portfolios, and actual returns

A distinct literature defines market-based variance directly from transaction data. For a static portfolio formed at $$VS(\tau;\Theta) = \frac{1}{\tau}\,\mathbb{E}_0^{\mathbb{Q}_\Theta}[QV(\tau)].$$6, observed trades in each constituent security over an averaging interval are rescaled so that the normalized traded volume in security $$VS(\tau;\Theta) = \frac{1}{\tau}\,\mathbb{E}_0^{\mathbb{Q}_\Theta}[QV(\tau)].$$7 sums to the portfolio holding $$VS(\tau;\Theta) = \frac{1}{\tau}\,\mathbb{E}_0^{\mathbb{Q}_\Theta}[QV(\tau)].$$8. Aggregating normalized values and volumes gives synthetic portfolio trades

$$VS(\tau;\Theta) = \frac{1}{\tau}\,\mathbb{E}_0^{\mathbb{Q}_\Theta}[QV(\tau)].$$9

so the portfolio can be analyzed as if it were a single traded security (Olkhov, 10 Apr 2025).

In that construction, the market-based variance of the portfolio return takes the form

$\tau \mapsto VS^{mkt}(\tau)$0

where $\tau \mapsto VS^{mkt}(\tau)$1 is the coefficient of variation of trade values, $\tau \mapsto VS^{mkt}(\tau)$2 is the coefficient of variation of trade volumes, and $\tau \mapsto VS^{mkt}(\tau)$3 is the normalized covariance between value and volume (Olkhov, 15 Oct 2025). If trade volumes are assumed constant, the decomposition collapses to the Markowitz expression. Without that assumption, the variance becomes a fourth-degree polynomial in the relative invested amounts, and its coefficients are not the covariances of security returns (Olkhov, 10 Apr 2025).

The same program has been extended from arbitrary portfolios to the market portfolio and to the entire market viewed as a synthetic security. In that setting, the return and variance of the held market portfolio need not equal the return and variance of the current market trading flow, because current trade shares $\tau \mapsto VS^{mkt}(\tau)$4 need not match the initial share fractions $\tau \mapsto VS^{mkt}(\tau)$5 (Olkhov, 15 Oct 2025). A further refinement expands the market-based variance around the Markowitz approximation as a Taylor series in the coefficient of variation of trade-volume fluctuations. The paper derives

$\tau \mapsto VS^{mkt}(\tau)$6

and shows limiting cases in which Markowitz variance may vastly undervalue or overestimate the portfolio variance and risks (Olkhov, 29 Jul 2025).

A closely related line studies “actual” returns of investors. There, trade-value-weighted averages and volatilities are defined for the realized return on a single sale, then for a single investor over a trading day, and finally across investors. At each level, volatility is written as a function of the moments, volatilities, and correlations of current and past trade values (Olkhov, 2023). This suggests that trade-based market variance is not only a portfolio-construction issue but also a cross-sectional statistic of realized investor outcomes.

5. Quadratic risk, martingale measures, and factor-based extensions

In incomplete-market hedging, market-based variance appears as the minimal variance of hedging error consistent with the traded market. In a model with external non-Gaussian OU factors, the variance-optimal martingale measure is constructed explicitly through a square-integrable density process, and the mean-value process of a claim is characterized by a BSDE with jumps. Combining the BSDE solution with the VOMM yields a globally risk-optimal mean-variance hedge (Dai, 2014). Here the relevant variance is neither a historical sample variance nor a purely statistical forecast; it is the minimal achievable quadratic hedging error under the market’s incomplete spanning structure.

In constrained multi-period mean-variance portfolio selection with dynamic factor models, the same dual viewpoint appears through the variance-optimal signed supermartingale measure. The optimal policy is a piecewise linear feedback in wealth, driven by the state-dependent processes $\tau \mapsto VS^{mkt}(\tau)$7 and $\tau \mapsto VS^{mkt}(\tau)$8, and

$\tau \mapsto VS^{mkt}(\tau)$9

The quantity $\tau \mapsto VIX^2(\tau)$0 is explicitly described as a market-based slope parameter of the mean-variance frontier (Gao et al., 25 Feb 2025).

A factor-risk extension appears in climate portfolio construction. When carbon risk is represented by a Brown-Minus-Green factor and time-varying carbon betas, the covariance matrix acquires an additional systematic term,

$\tau \mapsto VIX^2(\tau)$1

In that setting, absolute carbon beta is a driver of variance, and minimum-variance portfolios tend to prefer low absolute carbon risk even without explicit climate constraints (Roncalli et al., 2021).

6. Limitations, misconceptions, and current interpretation

The literature does not use “market-based variance” for a single invariant object. Depending on context, it may mean a risk-neutral variance term structure, a contract-level payoff variance, a transaction-based portfolio variance, or a quadratic-risk quantity defined through a variance-optimal pricing measure. This suggests that the phrase denotes a family of market-anchored variance concepts rather than one canonical statistic.

Several recurrent misconceptions are addressed directly in the cited work. First, $\tau \mapsto VIX^2(\tau)$2 is not literally the variance swap rate in jump models; using VIX as a variance-swap proxy can severely weaken identification in calibration (Yoo, 9 Sep 2025). Second, in the trade-based portfolio literature, Markowitz variance is not presented as the general formula but as the constant-volume approximation; once consecutive trade volumes are random, higher-order terms appear and the Markowitz estimate may be either too low or too high (Olkhov, 15 Oct 2025). Third, in option pricing, mean value alone is not an adequate description of risk: PEW and payoff variance provide distinct information about total-loss probability and payoff dispersion (Ben-Meir et al., 2012).

A further empirical caveat concerns volatility derivatives. Daily SPX evidence shows that non-linearities in the spot–volatility relationship have little impact on at-the-money volatility dynamics and on the skew–stickiness ratio, but can have a significant effect on the pricing and hedging of volatility derivatives, especially through the volatility of annualized variance and the behavior of short-dated variance exposures (Ségonne, 2015).

The most durable interpretation is therefore methodological. Market-based variance is variance inferred from market structure itself: option prices, variance-swap quotes, VIX futures, or actual trade values and volumes. Whether the objective is calibration, hedging, portfolio construction, or risk measurement, the defining claim is that variance should be computed from the same market information that supports pricing and trading, rather than being imposed solely as an exogenous historical statistic.