Machine-intelligence-for-finance

Measure how often the trial selected in-sample falls into the lower half of the out-of-sample ranking.

Separate noisy covariance eigenvalues from stronger directions and examine the effect on portfolio stability.

Train predictions for the quality of downstream portfolio decisions and test choices under uncertainty.

Represent parabolic equations through stochastic paths and train neural approximations against terminal conditions.

Learn an adapted hedging policy under transaction costs and compare its terminal risk with declared alternatives.

Compare correlation distance with information measures that can detect nonlinear dependence.

Compare clock, tick, volume, dollar, imbalance, and run sampling of the same transaction record.

Compare impurity, permutation, clustered, and orthogonalized importance when predictors share information.

Detect price events, assign first-passage labels, and compare meta-labeling with trend scans.

Use a fractional difference order to reduce persistence while retaining more memory than an ordinary difference.

Represent chronological dependence, train a TimeGAN model, and compare held-out real and generated sequences.

Build an asset hierarchy from correlations and allocate risk recursively without inverting the full covariance matrix.

Separate parameter search from final evaluation and examine whether the selected setting is stable through time.

Estimate trading frictions and volatility from observed prices when quotes or trade directions are unavailable.

Describe what is known at each date, distinguish statistical targets from estimates, and measure sampling uncertainty.

Optimize within clusters before combining them and compare nested allocations under estimation error.

Calculate mutually consistent payments in a liability network and examine how structure changes shock losses.

Measure event concurrency, calculate average uniqueness, and sample observations with less duplicated information.

Study the mean-variance frontier and measure how estimation errors move weights and realized risk.

Compare portfolio concentration through ordered partial sums and mixing by doubly stochastic matrices.

Translate calibrated signals into positions and examine costs, drawdowns, and strategy-risk limits.

Study how correlated errors limit an ensemble and whether predictor clusters remain stable across samples.

Separate variation within one probability law from disagreement across several supplied laws.

Compare hidden-state probabilities, change alarms, and refit policies while preserving chronological order.

Combine deterministic parallel calculations, named methods, chronological selection, and a checked portfolio result.

Read the public library as a software system: its modules, component selection, and Python-Julia design choices.

Account for non-normal returns, serial dependence, and the number of tried strategies when interpreting a Sharpe ratio.

Use recursive right-tailed tests and change-point methods to locate unusual persistence and level shifts.

Create controlled return systems with known covariance, drift changes, and repeated-sample uncertainty.

Fit threshold exceedances, compare downside-risk measures, and test a sequence of held-out risk forecasts.

Preserve time order and remove overlapping information when evaluating financial predictions.

Construct many time-aware validation paths while controlling overlap between financial labels.