ClassicalVolatilityModelBase<T>
Base Classes · AiDotNet.Finance.Volatility
Base for CLASSICAL (econometric) conditional-volatility models — GARCH and its variants — estimated by
maximum likelihood, exactly as in their original papers, rather than by neural-network training.
Mirrors how RegressionBase gives classical regressions the full model surface, but for the
volatility interface: it implements all of IVolatilityModel generically (forecast,
realized-vol, covariance/correlation, serialization, parameters) and delegates only the model-specific
conditional-variance recurrence + parameter set to the concrete subclass.
Implements: IFeatureAware, IVolatilityModel<T>
Known derived types: EGarchModel<T>, Garch11Model<T>, GjrGarchModel<T>
How It Works
Estimation follows the arch package’s defaults: a constant mean (ConstantMean), so the
parameter vector is [mu, volatility parameters...]; the variance recursion runs on the residuals
r - mu and is seeded with arch’s backcast, an exponentially weighted average of the first
min(75, n) squared residuals with weights 0.94^i; and the Gaussian quasi-log-likelihood is
maximised with a derivative-free Nelder–Mead simplex over the (constrained, via transform) parameters.
All arithmetic is generic in T and runs as whole-tensor engine operations.
Properties
| Property | Summary |
|---|---|
ILossFunction<T> DefaultLossFunction { get; } |
|
bool SupportsParameterInitialization { get; } |
|
bool SupportsTraining { get; } |
|
bool UseNativeMode { get; } |
|
int NumFeatures { get; } |
|
int ParameterCount { get; } |
Number of free parameters, including the constant mean. |
int PredictionHorizon { get; } |
|
int SequenceLength { get; set; } |
|
string ModelName { get; } |
Human-readable model name (e.g. |
Methods
| Method | Summary |
|---|---|
Dictionary<string, T> GetFeatureImportance() |
|
Dictionary<string, T> GetFinancialMetrics() |
|
Dictionary<string, T> GetVolatilityMetrics() |
|
IEnumerable<int> GetActiveFeatureIndices() |
|
IFullModel<T, Tensor<T>, Tensor<T>> Clone() |
|
IFullModel<T, Tensor<T>, Tensor<T>> DeepCopy() |
|
IFullModel<T, Tensor<T>, Tensor<T>> WithParameters(Vector<T> parameters) |
|
ModelMetadata<T> GetModelMetadata() |
|
ModelOptions GetOptions() |
|
T ForecastAnnualizedVol(Vector<T> returns, double periodsPerYear) |
One-step-ahead ANNUALIZED volatility forecast = √(variance × periodsPerYear). |
T ForecastNextVariance(Vector<T> returns) |
One-step-ahead conditional VARIANCE forecast given the observed return history. |
Tensor<T> CalculateRealizedVolatility(Tensor<T> highFrequencyReturns) |
|
Tensor<T> ComputeCorrelationMatrix(Tensor<T> returns) |
|
Tensor<T> ComputeCovarianceMatrix(Tensor<T> returns) |
|
Tensor<T> EstimateCurrentVolatility(Tensor<T> recentReturns) |
|
Tensor<T> Forecast(Tensor<T> input, double[] quantiles) |
|
Tensor<T> ForecastVolatility(Tensor<T> historicalReturns, int horizon) |
|
Tensor<T> Predict(Tensor<T> input) |
|
Vector<T> GetParameters() |
|
Vector<T> SanitizeParameters(Vector<T> parameters) |
|
bool IsFeatureUsed(int featureIndex) |
|
byte[] Serialize() |
|
void Deserialize(byte[] data) |
|
void Dispose() |
|
void FitReturns(Vector<T> returns) |
Fits the model to a return series by maximum likelihood (Nelder–Mead on the Gaussian QMLE). |
void LoadModel(string filePath) |
|
void LoadState(Stream stream) |
|
void SaveModel(string filePath) |
|
void SaveState(Stream stream) |
|
void SetActiveFeatureIndices(IEnumerable<int> featureIndices) |
|
void SetParameters(Vector<T> parameters) |
|
void Train(Tensor<T> input, Tensor<T> expectedOutput) |