Mispricing Process Dynamics

• Mispricing process dynamics is the study of deviations between observed asset prices and fundamental values, revealing effects of asymmetric information, market frictions, and agent heterogeneity.
• It leverages stochastic calculus, agent-based modeling, and Bayesian filtering to quantify mean reversion and feedback mechanisms in financial markets.
• Empirical validation shows that these models capture fat tails, volatility clustering, and regime-switching behaviors observed in real market data.

A mispricing process characterizes the temporal evolution of the deviation between observed asset prices and their fundamental values or efficient prices. The dynamics of mispricing encapsulate the effects of asymmetric information, agent heterogeneity, market frictions, and endogenous feedback mechanisms. Recent literature integrates stochastic calculus, agent-based modeling, and Bayesian filtering to rigorously specify and analyze mispricing processes under various market microstructures including risk aversion, slow-agent feedback, and informational frictions.

1. Formal Definitions and Canonical Models

The mispricing process $M_t$ is most generally defined as the difference between the observable market price $P_t$ and a theoretical reference price: either the asset's fundamental value $V_t$, an efficient (information-impounding) price $S_t$, or a model-based equilibrium price $P^\text{eq}_t$:

• $M_t = P_t - V_t$ (fundamentalist approach) (Gardini et al., 2024, Danilova, 2016, Bouchaud, 2010)
• $M_t = P_t - S_t$ (efficiency/efficient-pricing) (Derchu, 2020)
• $M_t = \log p_t - \log p^\text{eq}_t$ (deviation from a reference GBM) (Lamba, 2010)

In models with asymmetric information, the mispricing process can be formalized as a mean-reverting Ornstein–Uhlenbeck process:

$dU_t = -\lambda U_t\,dt + \sigma_m\,dW_t$

where $\lambda>0$ determines the speed of mean reversion, and $W_t$ is a Brownian motion independent of the permanent price driver (Buckley et al., 2011). In agent-based models, the mispricing process is derived from the excess demand or feedback dynamics imposed by heterogeneous trading rules, notably incorporating memory, bounded rationality, or behavioral distortions (Lamba, 2010, Kato, 2013, Gardini et al., 2024).

2. Stochastic Dynamics under Asymmetric Information and Risk Aversion

A principal result for markets with risk-averse market makers and informed traders is that the equilibrium price process resides in the Conditional Martingale of Marginal Variance (CMMV) class. A process $S_t$ is a CMMV if, under some equivalent martingale measure $\mathbb{Q}$, it can be represented as:

$S_t = S_0 + \int_0^t \sigma_s dW_s^{\mathbb{Q}}$

with

$$\mathbb{E}^\mathbb{Q}[(S_T - S_t)^2 | \mathcal{F}_t] = \int_t^T \sigma_s^2 ds, \quad \forall 0 \leq t \leq T \leq 1$$

The map $S_t = f(B_t, t)$, with $f$ solving the backward heat equation $f_t + \frac{1}{2}f_{xx}=0$, yields an increasing Markovian price process where private information is gradually impounded and mispricing asymptotically vanishes as $t \to 1$ (Meyer et al., 2017).

Risk aversion impacts mispricing predominantly through the drift in the real-world (physical) measure, via a Radon–Nikodym derivative change. The volatility term $\sigma_t = f_x(B_t, t)$ remains unaffected by risk aversion under the martingale measure but the drift is amplified for larger risk-aversion penalties, increasing adverse-selection effect and the expected “price-impact” penalty during imbalanced trades (Meyer et al., 2017).

3. Endogenous Dynamics, Heterogeneity, and Feedback

Mispricing processes are profoundly shaped by endogenous feedback mechanisms and agent heterogeneity:

• Agent-based models introduce slow agents who act on threshold-based rules, behavioral biases (herding, anchoring), and market depth that endogenously generates fat tails, volatility clustering, and cascades in price returns (Lamba, 2010).
• The deviation from the geometric Brownian motion baseline is entirely attributable to endogenous slow-agent feedback, with mispricing evolving as:

$M(n) = \log p(n) - \log p_{\text{eq}}(n)$

where the sole driver is the change in aggregate sentiment $\kappa \Delta \sigma(n)$. For realistic parameterizations, persistent mispricing, characterized by excess kurtosis (3–10) and power-law decay of volatility autocorrelations, emerges once herding exceeds the empirical instability threshold (Lamba, 2010).

• Feedback loops (e.g., spread-volatility loop) amplify mispricing and induce micro-liquidity crises where small shocks can be greatly magnified under conditions of low revealed liquidity, decoupling price from fundamental value on short/intermediate time scales (Bouchaud, 2010).

4. Bayesian Filtering and Learning in Price Formation

Mispricing in the Bayesian filtering framework is formalized as the deviation between the observable mid-price $P_t$ and the unobserved efficient price $S_t$, with $M_t = P_t - S_t$ (Derchu, 2020):

• The posterior on $S_t$ evolves via Zakai or Kushner–Stratonovich stochastic PDEs as trade arrivals incrementally reveal information.
• Instantaneous order arrival intensities are exponential in the spread and therefore in $M_t$:

$$\lambda_t^{a} = \lambda_0 e^{-a(\delta + M_t)}, \quad \lambda_t^{b} = \lambda_0 e^{-a(\delta - M_t)}$$

• In the linear–Gaussian regime with small spreads, the coupled dynamics are captured by SDEs for the posterior mean $x_t$ and variance $\sigma_t^2$:

$$dM_t = \sigma_t^2 a (dN_t^a - dN_t^b) - \sigma\,dW_t$$

where the jump–diffusion structure enables dynamic correction of mispricing by both trades and random diffusion of $S_t$.

Meta-order splitting and the resultant market impact are typically concave in meta-order size and time, with the mispricing process encoding the transient and permanent impact effects via integrals over the state-dependent volatility $\sigma_t^2$ and order flow (Derchu, 2020).

5. Market Frictions, Constraints, and Reflection Effects

Market frictions such as short-sale or budget constraints systematically skew the mispricing process through oblique-reflected diffusions in log-price space (Kato, 2013):

• Under short-sale prohibition, the reflected SDE ensures that $M_t = X_t - X_t^0 \geq 0$, representing systematic overpricing versus the frictionless case.
• With budget constraints, $M_t \leq 0$ and underpricing prevails.
• The reflected SDE framework extends the classical agent-based equilibrium, with the regulator process $L_t$ controlling the enforced nonnegativity/isolation of constrained states.

Long-run distributions inherit this skewness, with stationary densities derivable from Fokker–Planck equations subject to oblique boundary conditions (Kato, 2013).

6. Nonlinear and Regime-Switching Dynamics

In chartist–fundamentalist models, the temporal evolution of mispricing $m_t = P_t - V_t$ is controlled by piecewise-linear difference equations:

$$m_{t+1} = \begin{cases} (1+b)m_t - b m_{t-1}, & |m_t| \leq h \ (1+b-c)m_t - b m_{t-1}, & |m_t| > h \end{cases}$$

with $b$ and $c$ encoding chartist and fundamentalist strength, and $h$ a mispricing entry threshold (Gardini et al., 2024).

Distinct dynamical regimes emerge:

• For weak fundamentalists, constant mispricing persists (Grossman–Stiglitz scenario).
• For moderate parameterizations, oscillatory or quasiperiodic mispricing arises on fundamentalist chartist interaction (bounded mean-reversion cycles).
• Exogenous shocks induce regime switching between these behavioral equilibria, and persistent memory via initial conditions or compounded deviations.

Regime transitions, neutrality lines, and attractor coexistence lead to rich, time-dependent patterns of mispricing absent in linear models (Gardini et al., 2024).

7. Statistical Properties and Empirical Validation

Key statistical signatures predicted by rigorous mispricing process models include:

• Fat tails and volatility clustering in returns, matching empirical kurtosis and power-law autocorrelations (Lamba, 2010, Bouchaud, 2010).
• Stationary variance and autocorrelation properties determined by model parameters (e.g., for OU mispricing, $\text{Var}(U_\infty) = 1/(2\lambda)$) (Buckley et al., 2011).
• Empirical distributions of realized mispricing and price dispersion that exhibit sharply Laplace character in the short run and lognormal behavior for longer horizons, as observed in large-scale transaction datasets (Kaldasch, 2015).

The model-based statistical predictions align with empirical measurements of price dispersion, excess kurtosis, and regime durations in real markets (Lamba, 2010, Kaldasch, 2015).

---

References:

(Meyer et al., 2017, Lamba, 2010, Derchu, 2020, Bouchaud, 2010, Danilova, 2016, Gardini et al., 2024, Kaldasch, 2015, Buckley et al., 2011, Kato, 2013)