/// <summary>
/// Calculates the sample variance of an array using a rolling window.
/// The first element (index = 0) of the output array is the variance of points [0] to [length - 1] in 'array',
/// the second element of the output array is the variance of points [1] to [length] in 'array', and so on.
/// </summary>
/// <param name="array">Array of data for which we are calculating the variance. For time series, the last element (index = n-1), is the most recent.</param>
/// <param name="decayFactor">In most applications, the decay factor is between 0 and 1. Weigth on the last element in array is 1.0, the 2nd to last element d, 3rd to last d^2, ...</param>
/// <param name="length">>Length of rolling window length. Must be less than or equal to the lenght of the array of data.</param>
/// <returns></returns>
public static double[] RollingVariance(double[] array, double decayFactor, int length)
{
if (array.Length < length)
{
throw new ArgumentException("Not enought points in array. Number of points in array must be greater than or equal to length.");
}
double s1 = GeometricSeries.SumOfGeometricSeries(decayFactor, length);
double s2 = GeometricSeries.SumOfGeometricSeries(decayFactor * decayFactor, length);
double c = 1 / (s1 * s1 - s2);
double a = s1 * c;
double z1 = 0.0, z2 = 0.0;
double d = 1.0;
for (int i = 0; i < length; i++)
{
z1 += d * array[length - 1 - i] * array[length - 1 - i];
z2 += d * array[length - 1 - i];
d *= decayFactor;
}
double dToLength = Math.Pow(decayFactor, length);
double[] vArray = new double[array.Length - length + 1];
vArray[0] = a * z1 - c * z2 * z2;
for (int i = 1; i < vArray.Length; i++)
{
z1 = decayFactor * z1 + array[length - 1 + i] * array[length - 1 + i] - dToLength * array[i - 1] * array[i - 1];
z2 = decayFactor * z2 + array[length - 1 + i] - dToLength * array[i - 1];
vArray[i] = a * z1 - c * z2 * z2;
}
return(vArray);
}
/// <summary>
/// Returns the sample variance of an array.
/// </summary>
/// <param name="array">Array of data for which we are calculating the variance. For time series, the last element (index = n-1), is the most recent.</param>
/// <param name="decayFactor">In most applications, the decay factor is between 0 and 1. Weigth on the last element in array is 1.0, the 2nd to last element d, 3rd to last d^2, ...</param>
public static double Variance(double[] array, double decayFactor)
{
if (array.Length == 1)
{
return(0.0);
}
int n = array.Length;
double v = 0.0;
double d = 1.0;
for (int i = 0; i < n; i++)
{
v += d * array[n - 1 - i] * array[n - 1 - i];
d *= decayFactor;
}
double m = Mean(array, decayFactor);
double s1 = GeometricSeries.SumOfGeometricSeries(decayFactor, n);
double s2 = GeometricSeries.SumOfGeometricSeries(decayFactor * decayFactor, n);
double a = s1 / (s1 * s1 - s2);
double b = s1 * a;
v = a * v - b * m * m;
//Occasionally floating-point error will cause v to be less than zero when it should be exactly zero or very close to zero.
if (v < 0)
{
bool allEqual = Tools.ArrayAllEqual(array);
return(allEqual ? 0 : double.Epsilon);
}
return(v);
}
/// <summary>
/// Hybrid VaR is based on historical data, but weights more recent data more heavily.
/// </summary>
/// <param name="returnArray">Historical returns from which VaR is to be calculated. The last value, with the highest index is assumed to be the most recent data point.</param>
/// <param name="windowLength">Length of the VaR window. The number of historical returns that will be used to calculate the VaR.</param>
/// <param name="confidenceLevel">VaR confidence level. 95% and 99% are typical values. If confidenceLevel is 95% then we expect 95% of days to be better (have a more positive return) than the VaR.</param>
/// <param name="decayFactor">For a decay factor d, the most recent data point has weight 1, the second d, the third d^2, ...</param>
public static double HybridValueAtRisk(double[] returnArray, int windowLength, double confidenceLevel, double decayFactor)
{
if (returnArray.Length < windowLength)
{
throw new ArgumentException(string.Format("returnArray does not have enough data points to calcualte VaR for nDay = {0}.", windowLength));
}
if (confidenceLevel < 0.50)
{
double[] negReturnArray = new double[returnArray.Length];
for (int j = 0; j < returnArray.Length; j++)
{
negReturnArray[j] = -returnArray[j];
}
return(-HybridValueAtRisk(negReturnArray, windowLength, 1.0 - confidenceLevel, decayFactor));
}
//initialize VaR to the minimum return
double var = double.MaxValue;
for (int j = returnArray.Length - windowLength; j < returnArray.Length; j++)
{
if (returnArray[j] < var)
{
var = returnArray[j];
}
}
//Loop through all data to find the minimum return, then the second worst, ...
//This is generally faster than sorting when confidenceLevel is high, becaus we only need to identify a few minimums.
double totalWt = GeometricSeries.SumOfGeometricSeries(decayFactor, windowLength);
double varLimitWt = (1 - confidenceLevel) * totalWt;
double cummWt = 0;
List <int> alreadyHaveMinIndexList = new List <int>();
for (int i = 0; i < windowLength; i++)
{
double min = double.MaxValue;
int minIndex = -1;
for (int j = returnArray.Length - windowLength; j < returnArray.Length; j++)
{
if (returnArray[j] >= var && returnArray[j] < min && alreadyHaveMinIndexList.Contains(j) == false)
{
min = returnArray[j];
minIndex = j;
}
}
alreadyHaveMinIndexList.Add(minIndex);
double wt = Math.Pow(decayFactor, returnArray.Length - minIndex - 1);
cummWt += wt;
if (cummWt >= varLimitWt)
{
break;
}
var = min;
}
return(-var); //By convention VaR is quoted as a positive value, even though it corresponds to a loss.
}
/// <summary>
/// Returns the sample covariance between two arrays.
/// Arrays should be of equal length, and contain more than one element.
/// </summary>
/// <param name="array1"></param>
/// <param name="array2"></param>
/// <param name="decayFactor">In most applications, the decay factor is between 0 and 1. Weigth on the last element in arrays is 1.0, the 2nd to last element d, 3rd to last d^2, ...</param>
public static double Covariance(double[] array1, double[] array2, double decayFactor)
{
if (array1.Length != array2.Length)
{
throw new ArgumentException("Arrays must be the same length");
}
if (Tools.ArrayAllEqual(array1) || Tools.ArrayAllEqual(array2))
{
return(0.0);
}
int n = array1.Length;
if (n == 1)
{
return(double.NaN);
}
double covar = 0.0;
double d = 1.0;
for (int i = 0; i < n; i++)
{
covar += d * array1[n - 1 - i] * array2[n - 1 - i];
d *= decayFactor;
}
double m1 = Mean(array1, decayFactor);
double m2 = Mean(array2, decayFactor);
double S1 = GeometricSeries.SumOfGeometricSeries(decayFactor, n);
double S2 = GeometricSeries.SumOfGeometricSeries(decayFactor * decayFactor, n);
double A = S1 / (S1 * S1 - S2);
double B = S1 * A;
covar = A * covar - B * m1 * m2;
return(covar);
}