In briefA study revisiting Bonart's explanation of the square-root impact law
SummaryThis paper examines theoretical explanations of the square-root impact law, which links price impact to the square root of traded volume. The authors reformulate Bonart's argument based on diffusive prices and white increments, emphasizing that the market must whiten each participant's stationary stream of metaorders rather than isolated ones. They build a multi-agent propagator model and note that its falsifiable prediction, that isolated metaorders should not follow the law, seems at odds with known empirical data.
In briefDSTNet: A Causal Multi-Horizon Forecasting Model for Financial Time Series
SummaryDSTNet is a forecasting model that keeps the recent evolution of wavelet filter-bank magnitudes from seven technical indicators over a twenty-day lookback, using only causal information. It outputs one, three, five, and ten day forecasts in one pass. Tested on seven equity indices and gold against nine learned baselines and a random-walk benchmark, it beats persistence only at longer horizons in MAE and MAPE, while one-day results are inconclusive. A downstream allocation test does not show an equity-timing advantage.
In briefResidual Learning for Deeper Neural Asset Pricing Models
SummaryThis paper applies residual learning to neural network asset pricing models, preserving and refining shallow benchmark models so that the networks can be made deeper. Out of sample, value-weighted long-short portfolios built from deep residual models achieve a Sharpe ratio of 2.07, versus 1.92 for the corresponding shallow models and 0.89 for deep feedforward models. The authors conclude that model depth provides additional economic value in asset pricing and that the approach can scale such models.
In briefRobust portfolio risk management using CVaR with a reward-penalty mechanism
SummaryThis paper proposes robust portfolio selection models that add a reward-penalty mechanism to portfolio management, assuming the joint loss distribution of risky assets is uncertain but lies within a multivariate distribution set. It derives an explicit closed-form expression for the worst-case conditional value-at-risk and uses it to find allocations that minimize this worst-case CVaR. Experiments with real market data suggest the models can outperform several existing related models, and that combining downside risk with portfolio loss can improve risk management and returns. The results also show a trade-off between raising expected return and controlling worst-case CVaR.
In briefExponential investors facing weakly mean-reverting prices
SummaryThis paper studies a continuous-time market where asset prices show weak (sublinear) mean reversion and a nonzero drift. For an investor maximizing expected exponential utility, the authors show the certainty equivalent grows as O(T^{2β+1}), where 0<β<1 is the mean-reversion strength. An explicit asymptotically optimal strategy is also provided.
