GraphQL Swagger UI ReDoc OpenAPI Spec

Analysis Configuration

Markets are deeply interconnected, as the 2008 Global Financial Crisis proved beyond doubt. A shock in one can ripple through others—stocks, indexes, even regions. This analysis quantifies those spillovers, predicts volatility, and maps how risk moves between assets.

Choose Your Own Analysis below to explore market dynamics, forecast volatility, and uncover contagion paths.

Define Data Source

Use synthetic data for testing or actual market data.

Date Range

Specify the time period (max 5 years).

Select Symbols

Define which symbols to analyze (max 5 symbols). Choose your preferred method for selecting symbols
Selection Method
Manual Symbol Entry
MSFT, AAPL, NEM.US (artificial symbol names)
10, 20, 30 (corresponding to symbols above)

Data Preprocessing

Data normalization addresses the scale disparities between different stocks. Without proper scaling, a $3,000 Amazon stock would mathematically dominate a $50 Ford stock in correlation and volatility calculations, creating misleading results. Normalization transforms all assets to comparable scales while preserving their relative movement patterns and volatility characteristics, ensuring each asset contributes appropriately to portfolio-level metrics and spillover analysis.
Standardize: Centers data around zero with unit variance. Best for most analyses as it preserves relative relationships while making different assets comparable.
Min-Max: Scales to 0-1 range. Use when you need bounded values or when extreme outliers are important.
No Scaling: Keeps original prices. Only recommended when analyzing assets with similar price ranges or when absolute price levels matter.

ARIMA Parameters

ARIMA models predict future prices by analyzing patterns in historical data. By examining past price movements, trends, and random fluctuations, ARIMA provides a data-driven forecast of potential future price trajectories. This enables users to anticipate market behavior and make informed decisions.
Limits: p ≤ 5, d ≤ 2, q ≤ 5, Forecast Steps ≤ 20
The AR component assumes today's price depends on recent past prices. Think of it as "momentum" - if a stock has been rising for several days, AR captures the tendency for that trend to continue. Higher p values (3-5) capture longer-term momentum patterns but risk overfitting to noise. Start with p=1-2 for most financial series, increase if you see strong persistence in price movements.
Differencing removes trends to make the data stationary (constant mean/variance over time). Most financial prices are "integrated" - they wander randomly and don't revert to a mean. d=1 means taking first differences (today's price minus yesterday's), which usually converts trending price data into stationary returns. Use d=0 for already-stationary data like volatility indices, d=2 rarely needed except for highly trending series. This analysis pipeline automatically differences prices to returns, so this is default to 0.
The MA component captures how today's price reacts to recent random shocks or surprises. Unlike AR which uses actual past prices, MA uses past forecast errors. This helps model temporary market overreactions or corrections. Higher q values (3-6) are useful for financial data since markets often show clustered volatility and mean reversion after shocks.
Number of time periods to forecast into the future. Higher values (10-20) provide more detailed forecast plots but may be less reliable. Use 5-10 for daily data, 10-20 for weekly data. More than 20 steps is rarely meaningful for financial forecasting.

GARCH Parameters

GARCH models delve into market volatility, quantifying how much prices swing up and down over time. By identifying periods of heightened uncertainty (e.g., during market crashes) versus stable periods, GARCH models are indispensable for risk management and strategic planning. They help users understand and prepare for potential market turbulence.
Limits: p ≤ 3, q ≤ 3
ARCH captures how today's volatility depends on recent large price movements. If yesterday had a big price swing, today's volatility will likely be elevated. p=1 is usually sufficient since volatility shocks decay quickly. Higher values (2-3) may help with extremely volatile assets like crypto or emerging market currencies.
GARCH adds persistence to volatility clustering - the tendency for volatile periods to persist over time. This captures the "volatility of volatility" phenomenon where market stress tends to continue for weeks or months. q=1 handles most cases; q=2 for assets with very persistent volatility clustering like VIX or commodity futures.
Financial returns exhibit "fat tails" - extreme movements occur more frequently than normal distribution predicts. Student's t-distribution accommodates this reality, providing better risk estimates during market stress. Use normal distribution only for very stable, low-volatility assets. Consider skewed-t for assets with asymmetric return patterns.

Spillover Analysis

Spillover analysis uncovers how shocks in one market or asset propagate to others. For instance, a crash in the US stock market might ripple through European or Asian markets. This analysis sheds light on global market interconnectedness and contagion risks, enabling better risk assessment and diversification strategies.
Limits: Forecast Horizon ≤ 10, Max VAR Lags ≤ 20, Granger Significance ≤ 1, Rolling Window ≤ 365
Quantifies how much variance in each asset comes from shocks to other assets versus its own innovations. Essential for portfolio risk management and understanding correlation breakdown during crises. Disable only when analyzing single assets in isolation.
Diebold-Yilmaz: Standard spillover methodology
Baruník-Křehlík: Frequency domain decomposition
Steps ahead: Forecast horizon for spillover analysis
Model selection: Maximum lags to consider for VAR model
P-value threshold: For Granger causality tests
Time-varying spillovers: Use rolling window analysis