ARMA-GARCH-COPULA model and financial time series case | with code data

Original link: http://tecdat.cn/?p=3385

Recently I was asked to write a survey on copulas for financial time series

A description of the various models is obtained from the read data, including some graphical and statistical output.

> oil = read.xlsx(temp,sheetName ="DATA",dec =",")
copy

We can then plot the three time series

1 1997-01-10 2.73672 2.25465 3.3673 1.5400

2 1997-01-17 -3.40326 -6.01433 -3.8249 -4.1076

3 1997-01-24 -4.09531 -1.43076 -6.6375 -4.6166

4 1997-01-31 -0.65789 0.34873 0.7326 -1.5122

5 1997-02-07 -3.14293 -1.97765 -0.7326 -1.8798

6 1997-02-14 -5.60321 -7.84534 -7.6372 -11.0549
copy

The idea is to use some multivariate ARMA-GARCH process here. The heuristic here is that the first part is used to model the dynamics of the time series mean and the second part is used to model the dynamics of the time series variance.

This paper considers two models

  • Multivariate GARCH process on ARMA model residuals (or variance matrix dynamics model)
  • Multivariate model (copula-based) on ARMA-GARCH process residuals

Therefore, different series will be considered here, obtained as residuals of different models. We can also normalize these residuals.

ARMA model

> fit1 = arima(x = dat [,1],order = c(2,0,1))
> fit2 = arima(x = dat [,2],order = c(1,0,1))
> fit3 = arima(x = dat [,3],order = c(1,0,1))
> m < - apply(dat_arma,2,mean)
> v < - apply(dat_arma,2,var)
> dat_arma_std < - t((t(dat_arma)-m)/ sqrt(v))
copy

ARMA-GARCH model

> fit1 = garchFit(formula = ~arma(2,1)+ garch(1,1),data = dat [,1],cond.dist ="std")
> fit2 = garchFit(formula = ~arma(1,1)+ garch(1,1),data = dat [,2],cond.dist ="std")
> fit3 = garchFit(formula = ~arma(1,1)+ garch(1,1),data = dat [,3],cond.dist ="std")
> m_res < - apply(dat_res,2,mean)
> v_res < - apply(dat_res,2,var)
> dat_res_std = cbind((dat_res [,1] -m_res [1])/ sqrt(v_res [1]),(dat_res [,2] -m_res [2])/ sqrt(v_res [2]),(dat_res [ ,3] -m_res [3])/ SQRT(v_res [3]))
copy

Multivariate GARCH model

The first model that can be considered is the multivariate EWMA of the covariance matrix,

> ewma = EWMAvol(dat_res_std,lambda = 0.96)
copy

volatility

> emwa_series_vol = function(i = 1){
+ lines(Time,dat_arma [,i] + 40,col ="gray")
+ j = 1
+ if(i == 2)j = 5
+ if(i == 3)j = 9
copy

implicit correlation

> emwa_series_cor = function(i = 1,j = 2){
+ if((min(i,j)== 1)&(max(i,j)== 2)){
+ a = 1; B = 9; AB = 3}
+ r = ewma $ Sigma.t [,ab] / sqrt(ewma $ Sigma.t [,a] *
+ ewma $ Sigma.t [,b])
+ plot(Time,r,type ="l",ylim = c(0,1))
+}
copy

A multivariate GARCH, i.e. BEKK(1,1) model, for example using:

> bekk = BEKK11(dat_arma)
> bekk_series_vol function(i = 1){
+ plot(Time, $ Sigma.t [,1],type ="l",
+ ylab = (dat)[i],col ="white",ylim = c(0,80))
+ lines(Time,dat_arma [,i] + 40,col ="gray")
+ j = 1
+ if(i == 2)j = 5

+ if(i == 3)j = 9

> bekk_series_cor = function(i = 1,j = 2){
+ a = 1; B = 5; AB = 2}
+ a = 1; B = 9; AB = 3}
+ a = 5; B = 9; AB = 6}
+ r = bk $ Sigma.t [,ab] / sqrt(bk $ Sigma.t [,a] *
+ bk $ Sigma.t [,b])
copy

Simulate residuals from univariate GARCH model

A first step might be to consider some static (joint) distribution of the residuals. The univariate marginal distribution is

Profile of edge density (obtained using bivariate kernel estimator)

It is also possible to visualize the copula density (some nonparametric estimates above, parametric copula below)

> copula_NP = function(i = 1,j = 2){
+ n = nrow(uv)
+ s = 0.3

+ norm.cop < - normalCopula(0.5)
+ norm.cop < - normalCopula(fitCopula(norm.cop,uv)@estimate)
+ dc = function(x,y)dCopula(cbind(x,y),norm.cop)


+ ylab = names(dat)[j],zlab ="copule Gaussienne",ticktype ="detailed",zlim = zl)
+
+ t.cop < - tCopula(0.5,df = 3)
+ t.cop < - tCopula(t.fit [1],df = t.fit [2])


+ ylab = names(dat)[j],zlab ="copule de Student",ticktype ="detailed",zlim = zl)
+}
copy

can consider this

function,

Compute empirical versions of the three sequences and compare them to some parametric versions,

>

> lambda = function(C){
+ l = function(u)pcopula(C,cbind(u,u))/ u
+ v = Vectorize(l)(u)
+ return(c(v,rev(v)))
+}
>

> graph_lambda = function(i,j){
+ X = dat_res
+ U = rank(X [,i])/(nrow(X)+1)
+ V = rank(X [,j])/(nrow(X)+1)

+ normal.cop < - normalCopula(.5,dim = 2)
+ t.cop < - tCopula(.5,dim = 2,df = 3)
+ fit1 = fitCopula(normal.cop,cbind(U,V),method ="ml")
d(U,V),method ="ml")
+ C1 = normalCopula(fit1 @ copula @ parameters,dim = 2)
+ C2 = tCopula(fit2 @ copula @ parameters [1],dim = 2,df = trunc(fit2 @ copula @ parameters [2]))
+
copy

But one might wonder whether the correlation is stable over time.

> time_varying_correl_2 = function(i = 1,j = 2,
+ nom_arg ="Pearson"){
+ uv = dat_arma [,c(i,j)]
nom_arg))[1,2]
+}
> time_varying_correl_2(1,2)

> time_varying_correl_2(1,2,"spearman")

> time_varying_correl_2(1,2,"kendall")
copy

Spearman and time-varying rank correlation coefficient

or Kendall correlation coefficient

For model relevance, consider the DCC model(S)

> m2 = dccFit(dat_res_std)
> m3 = dccFit(dat_res_std,type ="Engle")
> R2 = m2 $ rho.t
> R3 = m3 $ rho.t
copy

To get some predictions, use e.g.

> garch11.spec = ugarchspec(mean.model = list(armaOrder = c(2,1)),variance.model = list(garchOrder = c(1,1),model ="GARCH"))
> dcc.garch11.spec = dccspec(uspec = multispec(replicate(3,garch11.spec)),dccOrder = c(1,1),
distribution ="mvnorm")
> dcc.fit = dccfit(dcc.garch11.spec,data = dat)
> fcst = dccforecast(dcc.fit,n.ahead = 200)
copy

This article is an excerpt from "R Language ARMA-GARCH-COPULA Model and Financial Time Series Cases"

Posted by Iank1968 on Wed, 01 Feb 2023 11:12:24 +1030