arXiv Science⌕ Search
Home/Archive/arXiv

arXiv · 2610.24558

Optimal Trading Filters and the Value of Predictive Signals

Matteo Smerlak
View original source ↗Read source document ↗Export metadata ↓

Abstract

The value of a forecast depends on its accuracy, but also on its actionability. In trading, an investor must translate return forecasts into positions while accounting for trading costs and the lingering impact of earlier trades. We solve the stationary quadratic trading problem with transient price impact using tools from prediction theory: Wiener-Hopf factorization gives the optimal policy as the composition of a conditional forecast and an explicit causal linear filter. We recover the Gârleanu-Pedersen aim-portfolio under exponential resilience and derive fractional trading filters under power-law impact decay, requiring memory beyond finite-dimensional LQR control. At fixed positive risk penalty, signals of equal instantaneous predictive strength have equal frictionless Markowitz value. With trading costs, net attainable value is determined by the mean square of the friction-weighted conditional forecast. Under power-law impact with exposure bleeding, we find an intermediate optimal mean-reversion rate for Ornstein-Uhlenbeck signals.

Explore related subjects

q-fin.MFmath.OCphysics.soc-phq-fin.PM

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Matteo Smerlak. 2026-10-17. Optimal Trading Filters and the Value of Predictive Signals. https://arxiv.org/abs/2610.24558

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Arbitrage-Aware Probabilistic Multi-Step Forecasting of Implied Volatility Surfaces: Joint Generation of Surface-Return Trajectories Using Latent Diffusion

Implied volatility surfaces summarise the option market and are central to many financial applications. Probabilistic forecasting of their future evolution requires modelling two-dimensional geometry, temporal dependence, and predictive uncertainty while preserving economic admissibility. We propose a conditional latent diffusion framework for generating joint 30-step trajectories of implied volatility surfaces and underlying returns. An arbitrage-aware autoencoder learns a surface representation, while diffusion captures the evolution. Evaluated on SPX surfaces, the framework generates realistic probabilistic multi-step scenarios. Its predictive mean also outperforms persistence beyond the shortest horizons.

q-fin.MF↗
arXiv2026

Probability-Weighting Functions for Normal Tempered-Stable Laws: Orthogonal Channels and Information-Theoretic Indices

A probability-weighting function (PWF) maps objective probabilities into decision weights. We construct PWFs from changes in the distribution of returns rather than from a parametric family: with prior and posterior return distributions $F_{prior}$ and $F_{post}$, the PWF is $w(p) = F_{post}(F_{prior}^{-1}(p))$. Both distributions are normal tempered-stable (NTS), and the posterior differs from the prior along one of three channels, each isolated with the variance held fixed: scale, skewness and tail thickness. Around a symmetric benchmark with excess kurtosis 3, a 40% change in scale moves cumulative probabilities by up to 7 to 8 percentage points, a skewness of $\pm 0.5$ by 2 percentage points, and a change of $\pm 0.3$ in the tail index by about 2 percentage points. The three channels leave distinct signatures: the deviation $w(p) - p$ changes sign once, twice and three times, respectively. Three information-theoretic indices (a Fisher-normalized logit shift, a signed Jensen-Shannon divergence and a log-odds elasticity) summarize the distortions and avoid the endpoint singularity of the classical ratio $(w(p)-p)/[p(1-p)]$. Applied to NTS parameters calibrated to S&P 500 ETF daily returns over 2015-2025, the same channels produce distortions of 2 to 7 percentage points. Code that reproduces every figure and number in the paper is provided.

q-fin.MF↗
arXiv2026

Bias correcting transaction cost models for characteristic function based estimators

Effective transaction cost is a necessary step for devising any competing financial trading strategy in the absence of quote data. Classically, Roll's model estimated the transaction cost using the first order autocovariance of the difference between daily transaction prices, while Hasbrouck's Bayesian approach used iterative posterior simulation. However, Roll's model fails when the first order autocovariance becomes positive, while Hasbrouck's estimator is computationally intensive, especially when many securities are involved. We develop empirical characteristic function methods with a first order bias correcting for spread estimation, and establish the asymptotic distribution of the corrected estimator under suitable conditions. Our numerical experiments illustrate the bias-variance tradeoff of characteristic-function methods compared to Roll's model, and their computational advantage over Hasbrouck's model.

q-fin.MF↗