Research
Every research paper we have written up, newest paper first. New arrivals are screened automatically for topical fit with systematic trading; the ones that clear it are read, written up, and — where the idea is testable on our data — backtested. Particular focus on alpha hypotheses that can be formalised and tested. The Radar also covers portfolio construction, market risk and execution where the research is directly relevant to systematic investment processes. Follow new entries by RSS.
16,915 papers screened · 588 on the radar · 24 shown
Language models that read financial news are usually trained on text from the same years used to test them, so a backtest can reward hindsight.
PAPER REPORTS · Monthly, price+text combination across all readers and models, 63-day shocks, test 2022–2024, net of 10bp per dollar… · Same monthly buffered portfolio: net return 12.4%/yr, volatility 10.4%, max drawdown 8%, turnover 60% per month,…
OUR BACKTEST · Sharpe -0.73 · Return -19.9% · Max DD -34.5%
A conditional normalizing flow can produce extreme scenarios even when its point forecasts appear reasonable. We compare two fitted models for daily Treasury yield changes at eleven maturities reconstructed from Federal Reserve curve parameters.
PAPER REPORTS · Stabilized flow, development windows (5×60 dates, 2002–2026; forecast metrics only, no trading strategy or costs):… · Stabilized flow, subsequent window April 13–July 8, 2026 (frozen checkpoints): scenario-mean MSE 0.3351, 90% coverage…
OUR BACKTEST · Sharpe 0.03 · Return +0.2% · Max DD -5.6%
Correlation matrices combine nonlinear positive-definiteness constraints with a simultaneous row-column permutation symmetry. Their generation therefore involves two distinct problems.
OUR BACKTEST · Sharpe 0.00 · Return +0.0% · Max DD 0.0%
Wavelet-based financial forecasters typically use the transform only to denoise, or reduce it to a single spectral snapshot at the forecast origin, and the convolution that produces the coefficients is usually bilateral, so it can read past the forecast…
PAPER REPORTS · S&P 500 DSTNet asymmetric risk-off allocation, test 2023-06-01 to 2024-05-31, 10 bps per unit turnover, idealised… · NYSE DSTNet risk-off, same period, 10 bps: Sharpe 1.80, MaxDD 5.65%, CumRet 11.31%, exposure 56.77%, total turnover…
OUR BACKTEST · Sharpe 0.46 · Return +21.1% · Max DD -24.7%
We introduce a deep kernel hedging framework that combines the flexibility of deep learning with the structural inductive bias of kernel methods.
PAPER REPORTS · S&P500 out-of-sample (Jan 1 2023 to May 31 2026), ATM European call, quadratic hedging with order-3 signature features,… · S&P500 out-of-sample (Jan 1 2023 to May 31 2026), ATM European call, CVaR hedging, frictionless.
OUR BACKTEST · Sharpe 2.30 · Return +22.0% · Max DD -2.8%
Predicting stock price movements is highly challenging due to the noisy, nonlinear, and multi-factor nature of financial markets.
OUR BACKTEST · Sharpe 0.70 · Return +20.9% · Max DD -6.5%
Hybrid Deep Learning for equity index forecasting is limited by three problems: propagation of OHLCV noise into derived technical indicators (TIs), channel-indiscriminate multi-scale decomposition that conflates heterogeneous frequency signatures, and static…
PAPER REPORTS · WaVeFuse directional strategy, 365-day 2023 out-of-sample, 10bp per-trade cost: mean CAGR 24.6%, mean Sharpe 3.69, mean… · Per-index directional (2023, 10bp): KOSPI CAGR 34.4%, Sharpe 4.73 [95% CI 4.37-5.09], max DD 1.9%; GDAXI 13.3%, Sharpe…
OUR BACKTEST · Sharpe 0.03 · Return +2.7% · Max DD -41.1%
This paper introduces FinAnalyzeNet, a novel hybrid deep learning framework designed to address the complex challenges of financial volatility prediction in dynamic market environments.
PAPER REPORTS · Sharpe ratio 1.42 for the volatility-based trading strategy (test period 2021-2023 implied; no transaction-cost… · Directional accuracy 84.7% (sd 1.2%) on volatility direction, test period 2021-2023
OUR BACKTEST · Sharpe 0.64 · Return +21.0% · Max DD -18.7%
MASTER's inter-stock multi-head attention -- the module responsible for modeling cross-sectional stock relationships -- accounts for 42.5% of model parameters and 25% of predictive value.
PAPER REPORTS · Long-short portfolio Sharpe at N~3,500 (full A-share, test 2019-2020, 4 seeds, frictionless): PureLSTM 5.6 +/- 0.5;… · IC / Rank IC as the primary reported performance: CSI300 Original 0.058 +/- 0.007 / 0.066 +/- 0.003 (5 seeds); CSI800…
OUR BACKTEST · Sharpe -0.58 · Return -50.8% · Max DD -63.3%
Algorithmic trading now represents a market exceeding $20 billion, where even marginal gains in signal robustness can translate into economically significant returns.
PAPER REPORTS · Hybrid ensemble, OOS 2025 (252 trading days), 2.2bp per trade per leg: total return 51.26%, Sharpe 2.44, Sortino 5.35,… · Hybrid ensemble, OOS 2025: annualised CAPM alpha 0.423 (p = 0.011), beta 0.048, Probabilistic Sharpe Ratio 0.960
OUR BACKTEST · Sharpe 0.75 · Return +696.5% · Max DD -365.1%
Implied volatility surfaces summarise the option market and are central to many financial applications.
PAPER REPORTS · Surface point-forecast RMSE aggregated over 30 horizons: 0.01262 vs persistence 0.01343, +6.09% gain; MAE gain +3.45%;… · h=1: RMSE 0.00619 vs persistence 0.00535 (-15.72%); MAE -34.49%
OUR BACKTEST · Sharpe -0.00 · Return -1.0% · Max DD -329.1%
Deep hedging is a data-driven approach to learn hedging strategies. It relies on synthetic price paths generator, as real market data is often limited for training.
OUR BACKTEST · Sharpe 1.78 · Return +5.9% · Max DD -0.6%
Abstract A key puzzle in finance is why algorithmic traders with advanced neural models sometimes fail to beat simple traditional strategies, while in other cases they clearly outperform them.
PAPER REPORTS · Cluster 0 (most efficient), test 2025-2026, net of 0.1% one-way costs: DDPG annualised return 34.72%, cumulative… · Cluster 0 best classical: HRP Sharpe 1.591, Calmar 2.3562, max drawdown -13.07%, annualised return 30.80%
OUR BACKTEST · Sharpe 0.78 · Return +25.3% · Max DD -11.7%
We develop a scalable adjoint-to-control framework for continuous-time portfolio choice under smooth pointwise constraints.
OUR BACKTEST · Sharpe 0.19 · Return +24.7% · Max DD -66.8%
Diffusion generative models have rapidly emerged as powerful tools for modeling complex financial data.
OUR BACKTEST · Sharpe 0.53 · Return +79.3% · Max DD -50.5%
Financial forecasting models are typically developed in full precision, yet production deployment often requires low-precision inference to reduce memory and computational cost. Post-training quantization (PTQ) enables such deployment without retraining.
PAPER REPORTS · FP32 mean daily IC (cross-sectional Spearman, averaged over test dates, 2018-2025 walk-forward): TSMixer 0.490±0.043,… · Baseline mean daily IC over the same folds: pooled HAR 0.408, persistence (trailing 5-day realized vol) 0.315
OUR BACKTEST · Sharpe 0.41 · Return +41.7% · Max DD -37.2%
Specialist training beats generalist scale when forecasting financial statements. To our knowledge, no prior work jointly forecasts complete financial statements beyond one year, yet in a discounted-cash-flow valuation most firm value sits past that window.
PAPER REPORTS · Forma change-space R^2 0.289 on full test sample, test period 2010-2024, no transaction costs applicable (forecasting… · Forma per-horizon R^2: 39.0% at h=1 falling to 22.5% at h=20 (full sample); 38.0% at h=1 to 23.7% at h=20 (LLM sample)
OUR BACKTEST · Sharpe 0.57 · Return +17.5% · Max DD -10.5%
Foundation models for time series forecasting demonstrate impressive zero-shot generalization but often underperform on specialized domains such as high-frequency finance.
PAPER REPORTS · Mean per-day correlation 0.3730 (GatedLinear+RF, 40 test days per stock, Dec 2024-Jan 2026 sample, transaction costs… · Pooled correlation 0.5972 and cross-day correlation 0.5631 (GatedLinear+RF, same 40-day test window)
OUR BACKTEST · Sharpe -0.42 · Return -0.6% · Max DD -0.8%
Predicting financial asset returns remains one of the most difficult challenges in empirical finance, driven by the low signal-to-noise ratio and the semi-strong form of market efficiency.
PAPER REPORTS · Sector LSTM (main 3-layer, k=10, 1995-2024): mean daily long-short return 0.100% before costs, Newey-West t = 7.81,… · Sector LSTM appendix figures (after 2bp per half turn): mean daily return 0.053%, t-statistic 4.106, annualized return…
OUR BACKTEST · Sharpe -0.03 · Return -9.6% · Max DD -140.3%
Control policies optimized in simulation can perform poorly in the real system when the parameters $x$ of the simulator are estimated from limited data but the resulting parameter uncertainty is not represented inside the simulation.
PAPER REPORTS · In-simulator spectral risk (x100, warm-up H=T), BS-VOL: RLM-trained policy 10.15 on RLM paths and 10.33 on SLM paths;… · In-simulator spectral risk (x100, warm-up H=T), HESTON-CORR: RLM policy 19.87 (SLM paths) / 19.93 (RLM paths) vs SLM…
OUR BACKTEST · Sharpe -0.18 · Return -66.9% · Max DD -81.1%
Financial sentiment classifiers are commonly evaluated against human labels, but strong linguistic performance does not necessarily imply economically useful return predictability. This study separates these questions through two experiments.
PAPER REPORTS · One-day mean rank IC, 2019 Benzinga S&P 100 sample: FinBERT 0.0143 (largest), Financial-RoBERTa 0.0141, Naive Bayes… · FinBERT one-day long–short: gross total return 12.96%, annualized Sharpe 1.11, max drawdown -6.37% (2019, gross — no…
OUR BACKTEST · Sharpe -1.16 · Return -61.4% · Max DD -64.2%
Local-stochastic volatility (LSV) combines vanilla marginals with richer smile dynamics, but calibration requires a slow, noisy and sequential McKean--Vlasov fixed point. We learn a projection-consistent operator for the calibration triple.
PAPER REPORTS · Calibration latency 0.60 ms/surface vs 98.5 ms particle baseline (paired, same hardware, synthetic held-out states) · Vanilla repricing RMSE 58.2 +/- 3.3 bps on 8 held-out surfaces, 2 seeds (spread across seeds, not a confidence…
OUR BACKTEST · Sharpe 0.53 · Return +5.6% · Max DD -2.0%
We propose an arbitrage-aware latent flow-matching framework for unconditional implied volatility surface generation.
OUR BACKTEST · Sharpe -0.06 · Return -1.2% · Max DD -5.1%
Heavy-tailed diffusion models replace Gaussian noise by a Gaussian variance mixture: denoising Levy probabilistic models (DLPM) take the mixing variables i.i.d. across coordinates, while Student-t EDM shares one mixing variable per sample.
OUR BACKTEST · Sharpe -0.33 · Return -3.3% · Max DD -5.8%