How to Analyze and Simulate High-Frequency Data: NIG Model and VG Model

In today's financial markets, data is king. High-frequency data, in particular, plays an increasingly important role in financial trading and market analysis. However, for most quantitative traders, especially beginners, the application of high-frequency data remains limited to simple algorithmic trading and backtesting. But there are many advanced uses of high-frequency data. By leveraging it, traders and analysts can accurately capture instant market fluctuations and predict short-term market trends. For example, by analyzing stock price movements over very short time intervals, one can make faster and more precise trading decisions. Today, we introduce two models commonly used in quantitative trading – the NIG and VG models – both of which are based on high-frequency data analysis to forecast future price paths.

1. Normal Inverse Gaussian (NIG) Model

The Normal Inverse Gaussian (NIG) model is a statistical model used to describe the volatility of financial asset prices, especially in the realm of high-frequency financial data. It belongs to a broader class of statistical models known as "exponential Lévy models." These models provide a way to model the jump behavior of financial asset prices – that is, sharp changes occurring over extremely short periods. The NIG model is valued for its ability to capture the "fat tails" and "peakedness" of financial asset returns, meaning that price changes are more extreme and more frequent than what a normal distribution would predict.

1.1 Model Formula

The formula for the Normal Inverse Gaussian (NIG) model is a mathematical expression describing the logarithmic change in asset prices. It can be represented as a stochastic process whose increments are defined by the following equation:

Xt=σW(τ(t))+θτ(t)+btXt​=σW(τ(t))+θτ(t)+bt

In this formula:

• XtXt​ represents the log-price at time tt.
• σσ denotes the standard deviation of price changes, controlling the volatility of the model.
• W(τ(t))W(τ(t)) is a Brownian motion (or Wiener process) under a random time or "stochastic clock", where τ(t)τ(t) is an independent stochastic process, typically following an Inverse Gaussian distribution.
• θθ and bb are linear trend terms; θθ affects the skewness (tilt) of log-returns, while bb is the drift or trend term for the log-price.
• τ(t)τ(t) is an Inverse Gaussian subordinator, which is the main distinction of this model from standard Brownian motion models (such as the Black-Scholes model).

1.2 Analytical Approach

The use of the Normal Inverse Gaussian (NIG) model in simulating and analyzing high-frequency financial data primarily involves fitting and evaluating intraday price movements. This model effectively captures the statistical properties of asset prices sampled at high frequencies, such as skewness and kurtosis of intraday data. The analytical steps are as follows:

• Model Setup: Define the NIG model for the logarithmic changes in asset prices, determining the key parameters including volatility (σσ), skewness control (θθ), and trend (bb).
• Parameter Estimation: Using high-frequency data (e.g., minute‑by‑minute or hourly trade data), estimate model parameters via Maximum Likelihood Estimation (MLE) or Method of Moments (MME). These methods adjust the model using observed price changes to best match real market behavior.
• Simulation and Validation: Once parameters are estimated, the model can be used to simulate price paths or perform other statistical analyses. For example, simulations can verify the model’s forecasting accuracy and check whether it captures key features such as sharp peaks and fat tails.
• Empirical Analysis: By analyzing tick data at different time intervals (e.g., every 20 minutes), researchers can evaluate the stability of parameter estimates. If estimates remain stable across a range of sampling frequencies, it indicates that the NIG model is suitable for that time scale.

2. Variance Gamma (VG) Model

The Variance Gamma (VG) model is a statistical model used in financial mathematics to simulate asset price movements. It is primarily employed to more accurately describe the statistical behavior of asset returns, especially extreme cases that deviate from normal volatility patterns.

The VG model builds upon classical Brownian motion (i.e., random walk models like ordinary stock price models) by incorporating additional randomness – a "time change" or "stochastic clock" – which allows it to better capture jumps and spikes observed in real financial data. Compared to traditional Brownian motion, the VG model offers a better fit for the fat‑tailed and peaked distributions seen in practice.

By providing a model for fat‑tailed distributions, the VG model helps risk managers assess and manage the risk of extreme market moves. In financial markets, the pricing of many complex derivatives (such as options and futures) relies on accurate predictions of future price volatility. Because the VG model can precisely describe price fluctuations, it is widely used for pricing such instruments.

2.1 Model Formula

The basic formula of the Variance Gamma model extends classical Brownian motion by introducing an independent random time process (or subordinator), enabling a more accurate description of jumps and extreme fluctuations in asset prices. Its general form is:

Xt=σW(τ(t))+θτ(t)+μtXt​=σW(τ(t))+θτ(t)+μt

where:

• XtXt​ is the log-price at time tt.
• σσ is the volatility parameter of asset returns, determining the magnitude of price fluctuations.
• W(τ(t))W(τ(t)) denotes a standard Brownian motion (Wiener process) evaluated at the random time τ(t)τ(t).
• θθ is the drift rate parameter, representing the average return rate of the asset price.
• μμ is a linear drift parameter, representing the constant part of the price trend.
• τ(t)τ(t) is an independent stochastic process, typically chosen as a Gamma‑distributed random process, used to model trading activity or random time variations in financial markets.

In the VG model, the introduction of the random time process τ(t)τ(t) allows the model to simulate sharp price movements over short intervals – something ordinary Brownian motion cannot achieve. In practice, the model's parameters are adjusted and estimated according to specific financial data for optimal fitting.

2.2 Application Steps of the Variance Gamma Model in High-Frequency Data Analysis

The following are the steps for processing high-frequency data using the VG model:

• Model Setup: Set the parameters of the VG model, including volatility (σσ), drift rate (θθ), and parameters of the random clock process. These parameters help the model adjust its behavior at different trading frequencies to reflect real market dynamics.
• Parameter Estimation: Use high-frequency data (e.g., minute‑by‑minute or second‑by‑second trade data) to estimate the VG model’s parameters. Common estimation methods include Maximum Likelihood Estimation (MLE) or Method of Moments (MME). These methods utilize observed price changes to calibrate the model to actual market behavior.
• Simulation and Validation: Once parameters are estimated, the model can simulate asset price paths or perform other statistical analyses. For instance, simulations can test the model’s predictive accuracy and check whether it captures key features such as spikes and jumps.
• Empirical Analysis: Analyze the model’s performance at different time intervals (e.g., every 20 minutes) to evaluate the stability of parameter estimates. If the estimates are stable across a range of sampling frequencies, it suggests the VG model is suitable for that time scale.

AllTick High-Frequency Data API

AllTick provides high-frequency data APIs that allow you to easily access a wealth of market data resources. Our data covers forex, Hong Kong stocks CFD, US stocks CFD, commodities, cryptocurrencies, and more. These interfaces are designed for exchanges, developers, quant teams, fintech companies, and professional institutions. They support all major client development languages and come with comprehensive documentation. We also offer free trials for various access tests.

If you are interested in our products, please contact our customer service via Telegram.