volarixs

volarixs - applied AI & ML to finance

Explore our latest posts on machine learning, market dynamics, strategy architecture and design

Feature Engineering
Jun 2, 2026

Shrinking the Feature Space: PCA & Autoencoders

Many features are redundant or noisy. High dimensionality = harder to generalize.

PCA
Autoencoders
Features
9 min read
Strategy
May 24, 2026

How Asset Managers Can Implement AI & Machine Learning

Part 2: Infrastructure, Governance & Roadmap. What it takes to implement AI in asset management.

AI Implementation
Governance
Roadmap
18 min read
Deep Learning
May 20, 2026

Neural Networks for Market Data: MLPs, CNNs & LSTMs

We are selective with deep learning. Expensive to train, easy to overfit, harder to debug.

Neural Networks
MLP
LSTM
12 min read
Research
May 14, 2026

Signal Half-Life and Decay: How Long Do ML Edges Really Last?

If you discover a signal today, how long will it work?

Signal Decay
Half-Life
Edge Persistence
13 min read
Strategy
May 7, 2026

How Asset Managers Can Use AI & Machine Learning in Investment Decisions

Part 1: Use Cases & Value. Real-world use cases: idea generation, regime analysis, risk management.

Asset Management
AI & ML
Use Cases
15 min read
Volatility
Apr 27, 2026

Modeling Market Turbulence: GARCH, EGARCH & HAR

Volatility ≠ returns: heavy tails, clustering, mean reversion. Dedicated volatility models are essential.

GARCH
EGARCH
HAR
10 min read
Time Series
Apr 9, 2026

ARIMA, SARIMAX & VAR: When Classical Time-Series Still Win

Explicitly model temporal dependence with transparent structure.

ARIMA
SARIMAX
VAR
9 min read
Benchmarks
Mar 31, 2026

Volatility Forecasting Benchmarks: GARCH, HAR, and ML

Compare GARCH, HAR, and ML models for volatility forecasting.

Volatility
GARCH
HAR
11 min read
Machine Learning
Mar 24, 2026

How Market Regimes Break ML Models

Financial machine learning rarely fails because the model is 'bad'. It fails because the market regime changed.

Regimes
ML
Backtesting
8 min read
Models
Mar 17, 2026

Boosted Trees for Alpha: XGBoost & LightGBM

Gradient boosting dominates tabular ML. Learn how XGBoost and LightGBM deliver strong performance.

XGBoost
LightGBM
Boosting
11 min read
Features
Mar 10, 2026

The 19 Most Important Features for Equity Return Forecasting

Most ML performance in finance doesn't come from the model — it comes from the features.

Features
Alpha
Equities
12 min read
Methodology
Feb 27, 2026

Rolling Windows for Financial ML: A Complete Guide

If you use financial data and your model does not use a rolling window, the backtest is wrong.

Rolling Windows
Time Series
Backtesting
10 min read
Evaluation
Feb 16, 2026

Beyond Sharpe: A Research Framework for Evaluating ML Trading Strategies

Sharpe ratio is dangerously incomplete for ML strategies.

Evaluation
Metrics
Sharpe
15 min read
Models
Jan 28, 2026

Random Forests in Finance: Nonlinear Signals Without the Drama

Tree-based ensembles capture nonlinearities and interactions in market data.

Random Forest
Extra Trees
Trees
10 min read
Models
Jan 5, 2026

From Linear Regression to Lasso: Fast, Interpretable Baselines

Linear and regularized regressions still do serious work in finance.

Linear Regression
Ridge
Lasso
12 min read
Regimes
Dec 12, 2025

Market Regimes, Clusters & HMMs: Teaching Models to Respect the Environment

Episodes where statistical properties are stable enough: high vol vs low vol, risk-on vs risk-off.

K-Means
GMM
HMM
11 min read
Architecture
Nov 23, 2025

Building a Universe-Wide Prediction Grid

An alpha factory needs predictions for every asset at multiple horizons from multiple models.

Prediction Grid
Scaling
Alpha Factory
14 min read
Evaluation
Oct 8, 2025

Regime-Conditioned Performance: Measuring ML Robustness

Most backtests report a single Sharpe. But ML models fail by regime.

Regimes
Robustness
Performance
12 min read
Deep Learning
May 20, 2026
12 min read

Neural Networks for Market Data: When to Use MLPs, CNNs & LSTMs

We are selective with deep learning. Expensive to train, easy to overfit, harder to debug. But they shine with rich feature sets and nonlinear temporal patterns.

1. Why We Are Selective with Deep Learning

Deep learning earns headlines, but on market data it carries real costs that a simpler model often avoids. Three in particular shape when we reach for it:

  • Expensive to train: requires significant compute resources
  • Easy to overfit on noisy financial signals
  • Harder to debug and explain than trees or linear models

2. Where They Shine

  • Rich feature sets (order book, microstructure)
  • Nonlinear temporal patterns
  • Large cross-asset datasets with shared structure

3. Types We Consider

MLP: simple feed-forward for tabular embeddings.

1D CNN / TCN: convolutions over time; good at local patterns.

LSTM / GRU: sequence models that retain state.

4. volarixs Stance

Neural Networks are one of the model classes you can select in an experiment, alongside regression, trees and boosted models, and time-series. We treat them as a second tier: fewer configurations, reached for on selected universes or factors once a simpler baseline has set the bar.

The discipline is to never run a deep model on its own. Each experiment is fit on the same walk-forward windows as its simpler counterparts and records train and test R² side by side, so a neural network has to clear a baseline on held-out data before it earns a place. Because regime context travels with every run, you can also ask the questions that matter most for deep models:

  • Does the out-of-sample fit actually beat a linear or boosted baseline?
  • Does it hold up across regimes, or only in the one it was trained in?
MLP
CNN
LSTM
Deep Learning

Get new research in your inbox

Applied AI & ML for the buy-side — new research on signals, regimes, and strategy design, straight to your inbox. No noise.

Ready to explore deep learning?

Build neural network models in volarixs.