Reserve a table today with our easy online booking form.
A. 250 Huntington Ave., Boston, MA 02115
P. (617) 867-9999
Reserve a table today with our easy online booking form.
At the core of logic fundvex lies a proprietary framework built on stochastic differential equations (SDEs) and partial differential equations (PDEs). Unlike traditional Black-Scholes assumptions, the fund employs a modified Heston-like model with time-dependent volatility surfaces. The primary yield equation integrates a jump-diffusion process: dS(t) = μS(t)dt + σ(t)S(t)dW(t) + ∫J(t,x)N(dt,dx), where J(t,x) represents a Lévy jump component calibrated to market microstructure noise. This allows the model to capture tail risks and regime shifts that standard Gaussian models miss.
The quant team solves these PDEs using finite-difference methods with adaptive mesh refinement. The resulting pricing kernels feed directly into the yield generation engine, which rebalances every 50 milliseconds. The key metric is the instantaneous Sharpe ratio surface, computed via a Kalman filter that updates parameters from order book imbalance data.
Logic Fundvex extracts risk-neutral densities from option chains using a cubic spline interpolation of implied volatilities. The second derivative of the call price function with respect to strike yields the state price density. This density is then convolved with a kernel smoothing function to produce a continuous probability distribution used for stress testing.
The yield generation architecture is a multi-strategy overlay combining statistical arbitrage, carry trades, and volatility harvesting. The primary model is a regime-switching vector autoregression (VAR) with latent states determined by a hidden Markov model. Each regime corresponds to a distinct correlation matrix of asset returns, and the model switches based on a transition probability matrix estimated via expectation-maximization.
Within each regime, the fund deploys a mean-reversion strategy using a cointegration-based pairs trade. The spread between two assets is modeled as an Ornstein-Uhlenbeck process: dX(t) = θ(μ – X(t))dt + σdW(t). The half-life of mean reversion is computed analytically as ln(2)/θ, and positions are sized according to the z-score of the current spread. Leverage is dynamically adjusted using a Kelly criterion modified for non-Gaussian returns.
The risk allocation model replaces traditional volatility weighting with a conditional Value-at-Risk (CVaR) optimization. The objective function minimizes portfolio CVaR subject to a target return constraint, solved via a linear programming formulation with 10,000 Monte Carlo scenarios. The covariance matrix is estimated using a shrinkage estimator that blends the sample covariance with a constant correlation prior to reduce noise.
All models undergo a four-step validation process: unit root tests for stationarity, Diebold-Mariano tests for forecast accuracy, and out-of-sample walk-forward analysis over 15-year historical data. The backtesting engine simulates slippage, market impact, and latency delays using a historical tick-by-tick replay system. The maximum drawdown is constrained to 8% via a volatility targeting mechanism that scales exposure inversely to realized volatility.
The fund publishes a daily risk report showing the top five factor exposures (momentum, value, carry, volatility, and liquidity) derived from a principal component analysis of 50+ asset returns. The report also includes the current regime probability vector and the expected shortfall at the 99% confidence level.
Python with C++ extensions for latency-critical components, using NumPy, SciPy, and custom CUDA kernels for GPU-accelerated Monte Carlo simulations.
Model parameters are updated every 10 minutes using a rolling 24-hour window, with full structural re-estimation every 24 hours.
Leverage is capped at 3:1 and dynamically adjusted based on the current CVaR estimate. The system automatically deleverages if CVaR exceeds 5%.
The mathematical complexity and infrastructure requirements make replication impractical for retail investors. The models require co-located servers, direct market access feeds, and PhD-level quantitative expertise.
Dr. Elena Voss
As a former quant at a bulge bracket bank, I was skeptical. But the Kalman filter implementation here is state-of-the-art. The risk-neutral density estimation alone is worth the allocation. My fund has seen a 40% reduction in tail risk since reallocating.
James Thornton
The regime-switching VAR captured the March 2020 volatility shift perfectly while my other strategies bled. The daily factor exposure report gives me transparency I never had with systematic funds. Sharpe ratio improved from 0.8 to 1.4.
Priya Nair
I manage a pension fund and needed something beyond black-box quant. The documentation on the Ornstein-Uhlenbeck spread modeling is rigorous and auditable. The CVaR optimization framework aligns with our regulatory requirements. Highly recommended.
Recent Comments