Predictive Investing.
Zero Stochastic Noise.

Next-generation quantitative engine combining Deep Learning, Markov Chain Regime detection, and Differentiable Optimization.

Live Financial News

Real-time market intelligence and global financial updates.

Finance News
Loading latest market news...

Submit Feedback

We build for you. Let us know what you need.

The Engine Room

Configure your matrix parameters and launch the optimization engine.

Quantitative Engine

Ask NOVA to Build Your Portfolio

Describe your ideal portfolio in plain English, and NOVA will automatically configure the matrix parameters below.

Matrix Configuration

5

Under the Hood: Active Mathematics

$$ \hat{y} = \sum_{k=1}^{K} f_k(X) + \lambda \|\beta\|_1 $$

Currently modeling predictive alpha via Gradient Boosted Decision Trees with L1-Norm feature selection penalization.

Saved Portfolios

Access and manage your historically optimized portfolio configurations.

📖 How to Use

This section stores your favorite portfolio configurations. Once you run an optimization in the main engine, simply click the green Save Configuration button above the report. Your asset weights, selected engine, and risk profile will be permanently stored here so you can deploy them later with a single click.

Backtest History

Review the historical out-of-sample performance of your past strategies.

📖 How to Use

Every time you execute the engine with Backtesting enabled, the full simulation history is recorded here. Use this dashboard to compare how different model configurations (like HMM vs Black-Litterman) performed on out-of-sample data over time. You can export these curves as CSV files for external analysis.

Execution Time Model Annualized Return Sharpe Ratio Action

Risk Dashboards

Advanced structural and macroeconomic risk analysis.

📖 How to Use

The Risk Dashboard automatically populates after you run a portfolio optimization. It provides deep diagnostic insights into the tail risks of your allocated weights, projecting expected Value at Risk (VaR) and applying historical stress scenarios (like the 2008 Financial Crisis) directly to your active configuration.

Tail Risk Breakdown (VaR & CVaR)

Run an optimization to view 95% and 99% VaR/CVaR metrics.

Macro Stress Scenarios

Run an optimization to view historical stress tests.

API & Webhooks

Integrate the Portfolio Engine directly into your automated trading systems.

📖 How to Use

This section is for developers running algorithmic trading pipelines. You can use the provided REST API endpoints to trigger portfolio rebalancing asynchronously. Configure a Webhook URL below, and the engine will automatically POST the final JSON allocation weights to your execution server the moment an optimization finishes.

REST API Access

Use your session token to access the engine programmatically.

POST /api/generate
Headers: X-Access-Key

Webhook Integration

Register a callback URL to receive asynchronous optimization results.

Quantitative Alpha Models

Our predictive engines represent the bleeding edge of quantitative finance. We deploy both deterministic mathematical equilibrium models and robust machine learning ensembles to forecast expected returns. Below is the strict architectural breakdown of every engine available to you.

1. CAPM (Capital Asset Pricing Model Prior)

The Core Mathematical Hypothesis

Established by William Sharpe in 1964, CAPM posits that investors should only be compensated for "systematic risk"β€”risk that cannot be diversified away. Expected returns are purely a function of an asset's covariance with the broader market baseline (its Beta).

$$ \mathbb{E}[R_i] = R_f + \beta_i(\mathbb{E}[R_m] - R_f) $$

When It Shines vs. Fails

✓ Exceptional in stable, highly correlated macro environments where Beta is a reliable, stationary predictor of upside.

✗ Breaks down structurally when idiosyncratic shocks occur (e.g. a specific sector collapsing while the index rallies) because it completely ignores individual asset momentum.

2. Global Pooled Panel Machine Learning (XGBoost & Lasso)

The Core Mathematical Hypothesis

Financial markets are deeply non-linear. This engine stacks gradient-boosted decision trees (XGBoost) to capture non-linear market regimes and anomalies, layered with a strictly penalized Lasso (L1 regularization) model to filter out extreme statistical noise. A Ridge Meta-Learner aggregates their predictions to ensure out-of-sample robustness.

$$ \hat{y} = \alpha \cdot \text{XGB}(x) + (1-\alpha) \cdot \text{Lasso}(x) $$ $$ \mathcal{L} = \text{MSE} + \lambda_1 \|w\|_1 $$

When It Shines vs. Fails

✓ Dominates in highly complex, multi-factor datasets where linear assumptions (like CAPM) fail to capture hidden Alpha.

✗ Machine Learning inherently risks overfitting; it can misallocate capital if the market shifts into an unprecedented, structurally unobserved macroeconomic regime (e.g. Black Swan events).

3. Black-Litterman (Market Equilibrium Prior)

The Core Mathematical Hypothesis

Pure historical optimization (Markowitz) leads to wildly concentrated portfolios that change radically day-to-day. Black-Litterman solves this using Bayesian shrinkage: it calculates the global market equilibrium (implied returns) as the mathematical "Prior", and updates it only if there is statistical evidence to deviate.

$$ \mathbb{E}[R] = \left[ (\tau \Sigma)^{-1} + P^T \Omega^{-1} P \right]^{-1} \left[ (\tau \Sigma)^{-1} \Pi + P^T \Omega^{-1} Q \right] $$

When It Shines vs. Fails

✓ The undisputed gold standard for generating robust, highly diversified portfolios that do not whip-saw across rebalancing periods.

✗ Inherently conservative. By anchoring to the market equilibrium, it may under-allocate to early-stage, explosive breakout trends compared to a pure momentum model.

4. Multifactor Regression (Fama-French + Momentum)

The Core Mathematical Hypothesis

An expansion of CAPM. Eugene Fama and Kenneth French proved that returns are not just driven by market Beta, but structurally driven by Small-Cap outperformance (SMB) and Value outperformance (HML). We append the Momentum factor (MOM) to ride prevailing trends.

$$ R_{i,t} - R_{f,t} = \alpha_i + \beta_1(R_{m,t} - R_{f,t}) + \beta_2 \text{SMB}_t + \beta_3 \text{HML}_t + \beta_4 \text{MOM}_t + \epsilon_{i,t} $$

When It Shines vs. Fails

✓ Extremely effective over long-term multi-year horizons, capitalizing on fundamental economic realities (value and size premiums).

✗ Vulnerable to "Factor Winter" where growth stocks dominate value for a decade (e.g. 2010-2020 Tech boom).

5. Bayesian Shrinkage (James-Stein Estimator)

The Core Mathematical Hypothesis

Historically estimated returns are riddled with estimation error. James-Stein shrinkage mathematically "shrinks" extreme individual asset estimates toward the grand mean of the portfolio. This is a statistical guarantee against trusting outlier data.

$$ \hat{\mu}_{\text{JS}} = w \cdot \mu_i + (1 - w) \cdot \mu_{\text{global}} $$where \( w \) is derived from variance of estimates.

When It Shines vs. Fails

✓ Reduces the impact of "noise" in the data, preventing the optimizer from chasing an asset just because it had a lucky 30-day run.

✗ Can shrink truly exceptional, paradigm-shifting assets back down to the average, muting potential asymmetric upside.

6. End-to-End Differentiable Optimization (SPO+)

The Core Mathematical Hypothesis

Traditional finance separates the prediction of returns from the optimization of weights. Predict-then-Optimize (SPO+) fuses them using a custom differentiable loss function. The machine learning model is trained to minimize the regret of the portfolio's actual performance, not just mean squared error of predictions.

$$ \mathcal{L}_{\text{SPO}+}(\hat{c}, c) = \max_{w \in \mathcal{S}} \left\{ (2\hat{c} - c)^T w \right\} - 2\hat{c}^T w^*(c) + c^T w^*(c) $$

When It Shines vs. Fails

✓ Maximizes true financial utility by directly penalizing bad allocations rather than just inaccurate forecasts.

✗ Computationally expensive. Requires calculating optimization gradients through a convex solver at every training step.

7. Regime-Adaptive Hidden Markov Model (HMM)

The Core Mathematical Hypothesis

Markets do not behave uniformly over time; they transition between hidden probabilistic states (Bull, Bear, Stagflation). The HMM assigns a probability that the market is in a specific latent state today, and dynamically adjusts the covariance and return expectations based strictly on the parameters of the detected regime.

$$ P(Z_t \mid X_{1:t}) \propto P(X_t \mid Z_t) \sum_{Z_{t-1}} P(Z_t \mid Z_{t-1}) P(Z_{t-1} \mid X_{1:t-1}) $$

When It Shines vs. Fails

✓ Instantly reallocates defensive assets when shifting from a high-growth to a high-volatility regime.

✗ "Whip-saw" risk. If the market falsely signals a crash regime but instantly rebounds, the model may trap the portfolio in cash.

Engine Documentation & Architecture

The complete pipeline architecture from web orchestration to stochastic optimization and execution. Below is the strict mathematical specification of our constraint matrices and data flow.

1. Dynamic GARCH(1,1) Volatility Modeling

Standard historical volatility treats risk as a flat, static number. This is mathematically flawed. Financial markets exhibit "volatility clustering"β€”large changes are followed by large changes. Our GARCH (Generalized Autoregressive Conditional Heteroskedasticity) overlay dynamically detects expanding volatility regimes and proactively shrinks risk exposures before tail events crush the portfolio.

$$ \sigma^2_t = \omega + \alpha r^2_{t-1} + \beta \sigma^2_{t-1} $$

This equation updates today's volatility (σ²_t) using yesterday's shock (α * r²_{t-1}) and yesterday's variance (β * σ²_{t-1}), creating a memory effect that standard deviation lacks.

2. L1-Norm Tax Optimization Penalty

High turnover strategies look great on paper but bleed capital to short-term capital gains taxes in the real world. When the Tax Optimization flag is toggled, we inject a strict L1-norm penalty vector directly into the convex solver. This forces the engine to mathematically weigh the theoretical alpha of a trade against the concrete tax drag of selling an existing position.

$$ \min_{w} \left( w^T \Sigma w + \lambda \| w - w_{\text{prev}} \|_1 \right) $$

The absolute difference penalty (L1) forces the optimizer to prefer zero-turnover unless the marginal return strictly exceeds the tax boundary Ξ».

3. Allocation Engine: CVaR vs. Risk Parity

CVaR (Conditional Value at Risk): Unlike Variance (which penalizes upside growth), CVaR explicitly isolates and minimizes only the worst 5% of historical tail crashes.

Hierarchical Risk Parity (HRP): Uses unsupervised machine learning (agglomerative clustering) to group correlated assets into sub-clusters, assigning capital purely based on risk distribution rather than unreliable return forecasts.

Exact Risk Parity (ERC): A non-linear root-finding algorithm that guarantees every single asset contributes the exact same marginal risk to the total portfolio variance, ensuring true diversification.

$$ \text{Risk Contribution}_i = \frac{w_i (\Sigma w)_i}{w^T \Sigma w} = \frac{1}{N} $$

4. HMM Macro Regime Detection

Financial markets operate in distinct probabilistic states (e.g., Bull, Bear, High-Inflation, Low-Growth). The engine uses a Hidden Markov Model (HMM) to classify the current state of the global economy in real-time. It then strictly filters historical covariances, ensuring the optimizer only trains on data from the current macro regime.

$$ A_{ij} = P(\text{Regime}_{t}=j \mid \text{Regime}_{t-1}=i) $$

How It Works: The Institutional Pipeline

Wealth Engine is not a simple calculator. It is a full-stack algorithmic trading pipeline. Below is the analytical breakdown of exactly how it transforms raw data into a strictly optimized portfolio.

Phase 1: Understanding the Current Economy

The engine first connects to live market data to pull historical prices and interest rates. However, assuming the market behaves the same way all the time is dangerous. Our engine uses an algorithm to classify the current "Regime" of the global economy (e.g., Bull Market vs. High Volatility Selloff). By understanding the current environment, the engine only focus on data that is relevant to today's reality, ignoring irrelevant past events. Whether analyzing standard equities, bond yields, or extreme crypto volatility, the regime-detection algorithms automatically adapt to structurally different asset classes.

$$ \text{Regime} = \arg\max_k P(Z_t = k \mid \text{Market Data}_{1:t}) $$

Phase 2: Forecasting Future Returns

Next, the engine needs to predict how much return each asset will generate. Instead of simple guessing, you choose a Quantitative Alpha Model. For example, if you choose our advanced ensemble, the engine calculates momentum, volatility, and market correlation for every asset. It then uses statistical learning to find patterns between these metrics and future returns, creating a precise mathematical forecast for your portfolio.

$$ \mathbb{E}[R_{t+1}] = f(\text{Features}_t, \text{Parameters}) $$

Phase 3: Risk Management & Allocation

With the forecasts in hand, the engine enters the Convex Solver. This is where it calculates the exact percentage weighting of each asset to maximize your returns while capping your downside risk. It does this by measuring how assets move together (Covariance) and explicitly minimizing the damage of the worst 5% of historical market crashes. If you enable Tax Optimization, it will actively avoid proposing trades that trigger short-term capital gains taxes.

$$ w^* = \arg\max_w \left( w^T \mathbb{E}[R] - \frac{\lambda}{2} w^T \Sigma w \right) $$

Phase 4: Stress Testing the Portfolio

Before delivering the final allocation, the engine attempts to "break" the portfolio. It runs thousands of randomized future simulations (Monte Carlo) to see how the portfolio handles uncertainty. It also tests the portfolio against a simulated 2008 Financial Crisis, the 2020 COVID-19 shock, and the 2022 Inflation/Rate shock. Only if the portfolio survives these rigorous thresholds does the engine finalize the output report.

$$ S_{k, t+1} = S_{k, t} \exp\left( \left(\mu - \frac{\sigma^2}{2}\right)\Delta t + \sigma \sqrt{\Delta t} Z \right) $$

Terms of Service

Please read these terms carefully before utilizing the Wealth Engine platform.

1. Not Financial Advice

This platform is an experimental mathematical research tool, not a registered financial advisor. The outputs, forecasts, and allocations provided by the engine are for informational and algorithmic research purposes only. Do not blindly risk personal capital based on the outputs of this engine without consulting a registered financial professional.

2. Historical Illusion

All optimizations and predictive models rely heavily on historical data backtesting. Past performance is definitively not indicative of future results. Market dynamics backtesting. Past performance is definitively not indicative of future results. Market dynamics can shift rapidly into unobserved regimes, rendering historical patterns obsolete. This is especially true for Cryptocurrencies, which exhibit extreme, non-stationary volatility.

3. Machine Learning Limitations

You are interacting with advanced predictive systems (NOVA, AETHELRED, VANGUARD). While these engines use state-of-the-art Deep Learning and Convex Optimization, financial markets are fundamentally chaotic systems. The engine is a sophisticated co-pilot, not a crystal ball. It is highly susceptible to "Black Swan" events and structural regime shifts.

4. No Liability

You are solely responsible for your own investment decisions. The creators, operators, and affiliates of Wealth Engine accept absolutely no liability for any financial losses, capital destruction, or damages resulting directly or indirectly from the use of this platform.

5. Institutional Access Key

Access to this platform is strictly limited to authorized users holding a valid institutional access key. You agree not to share, distribute, or otherwise compromise your access key or associated username. We reserve the right to revoke access at any time without notice.

Speed Trading Demo

Experience microsecond-level market simulation.

Why use this?

This tool simulates extremely fast trading conditions (High-Frequency Trading) allowing you to see how quickly market orders interact. It helps you understand how tiny price differences are captured in milliseconds before the rest of the market reacts. It's a risk-free way to watch high-speed algorithms in action.

Price & PnL

Simulation Metrics

Run a simulation to view metrics.

Limit Order Book (LOB) Dynamics

Pairs Trading

Discover mathematically linked assets to trade market neutrality.

What is Pairs Trading?

Pairs trading identifies two assets that historically move together. When their prices temporarily split apart, you can buy the underperforming one and sell the outperforming one, profiting when they inevitably snap back into alignment. This strategy is immune to overall market crashes!

Digital Asset Arbitrage

Automatically find price differences across crypto exchanges.

How to use this scanner

Crypto prices vary slightly between different exchanges like Binance and Kraken. This bot scans multiple global exchanges simultaneously to find instant, risk-free profit opportunities by buying low on one exchange and selling high on another, automatically factoring in all trading fees.

Engine Speed Test

Compare our core algorithms in Python vs compiled C++.

Why speed matters

In finance, complex calculations like Monte Carlo simulations can take hours. We wrote our core math engines in highly optimized C++ to run thousands of times faster than standard Python, delivering institutional-grade analytics to your browser almost instantly.

Run real-time benchmarks comparing the pure-Python implementation against the pybind11/Eigen C++ extensions.