TENET-CODE_TENET模型经典代码_

[<img src="https://github.com/QuantLet/Styleguide-and-FAQ/blob/master/pictures/banner.png" width="880" alt="Visit QuantNet">](http://quantlet.de/index.php?p=info) ## [<img src="https://github.com/QuantLet/Styleguide-and-Validation-procedure/blob/master/pictures/qloqo.png" alt="Visit QuantNet">](http://quantlet.de/) **TENET_SIM** [<img src="https://github.com/QuantLet/Styleguide-and-Validation-procedure/blob/master/pictures/QN2.png" width="60" alt="Visit QuantNet 2.0">](http://quantlet.de/d3/ia) ```yaml  Name of QuantLet: TENET_SIM Published in: TENET Description: 'estimates Conditional Value at Risk (CoVaR) of 100 financial institutions by using Single-Index Model with variable selection. The 110 covariates include log returns of 99 firms (except for firm k) 7 macro state variables and 4 firm k’s characteristics. Then generates the necessory files for other TENET quantlets. The data is not publicly published.' Keywords: 'tail, quantile regression, CoVaR, systemic Risk, variable selection, dimension reduction, risk, bic' See also: 'quantilelasso, SIMqrL1, TENET_Linear, TENET_total_connectedness, TENET_total_in_out_groups, TENET_group_network, TENET_total_in_out_individual, TENET_SIFIs, TENET_VaR_CoVaR' Author: Weining Wang, Lining Yu Submitted: Datafile: '100_firms_returns_and_macro_2015-04-15.csv, Bal_sheet.csv, VaR_movingwindows_20150617.csv' Input: - yw : (ws+1) response vector - xxw : px(ws+1) covariate matrix - tau : scalar quantile level - VaRM_est : p estimated VaR and macro variables Output: - lambda_sim[l] : scalar estimated penalization parameter - beta_sim[l, ] : p estimated coefficients - CoVaR_sim[l] : scalar estimated CoVaR - first_der[l] : scalar estimated first derivative - partial_der[l, ] : p estimated partial derivatives ``` ```r # Step 1: The main code of TENET based on quantile regression for Single-Index # Model with Variable selection technique # clear all variables rm(list = ls(all = TRUE)) graphics.off() # set the working directory # setwd('C:/...') # install and load packages libraries = c("quantreg", "KernSmooth", "SparseM", "MASS") lapply(libraries, function(x) if (!(x %in% installed.packages())) { install.packages(x) }) lapply(libraries, library, quietly = TRUE, character.only = TRUE) source("TENET_SIMqrL1.r") source("quantilelasso.r") # read the file which includes log returns of 100 firms and 7 macro state # variables x0 = read.csv("100_firms_returns_and_macro_2015-04-15.csv", header = TRUE) # all firms' characteristics from balance sheet information of 100 firms Bal_sheet_full = read.csv(file = "Bal_sheet.csv") # 7 macro state variables m = as.matrix(x0[, 102:108]) # estimated Value at Risk of 100 firms VaR = as.matrix(read.csv("VaR_movingwindows_20150617.csv")[-1])[, 1:100] # log returns of 100 firms xx0 = x0[, 2:101] # start the linear quantile lasso estimation for each firm for (k in 1:100) { cat("Firm:", k) # log return of firm k y = as.matrix(xx0[, k]) # log returns of firms except firm k xx1 = as.matrix(xx0[, -k]) # 4 firm characteristics from balance sheet informaiton of firm k BS = Bal_sheet_full[, (4 * k - 3):(4 * k)] # combine macro state variables and 4 firm characteristics MB = cbind(m, BS) # number of rows of log return n = nrow(xx1) # number of covariates p = ncol(xx1) + ncol(MB) # estimated Value at Risk of firms except firm k V = as.matrix(VaR[, -k]) # quantile level tau = 0.05 # moving window size equals 48 corresponds to one year weekly data ws = 48 lambda_sim = matrix(0, (n - ws), 1) beta_sim = matrix(0, (n - ws), p) covar_sim = matrix(0, (n - ws), 1) first_der = matrix(0, (n - ws), 1) partial_der = matrix(0, (n - ws), p) for (l in 1:(n - ws)) { print(l) yw = y[l:(l + ws)] MBw = MB[l:(l + ws), ] mb = matrix(0, ws + 1, ncol(MB)) # standardize macro state variables and 4 firm characteristics for (j in 1:ncol(MB)) { mb[, j] = (MBw[, j] - min(MBw[, j]))/(max(MBw[, j]) - min(MBw[, j])) } mb[is.na(mb)] = 0 MBw[is.na(MBw)] = 0 xx = xx1[l:(l + ws), ] # all the independent variables xxw = cbind(xx, mb) VaRM_est = as.numeric(c(V[l, ], mb[(ws + 1), ])) fit = sim(yw, xxw, tau, Qmaxiter = 2, l, k, LVaRest = VaRM_est) beta_sim[l, ] = fit$beta_final lambda_sim[l] = fit$lambda.fi # the final estimated CoVaR covar_sim[l] = fit$a.fi first_der[l] = fit$b.fi # the estimated partial derivatives partial_der[l, ] = fit$c.fi } write.csv(lambda_sim, file = paste("lambda_sim_", k, ".csv", sep = "")) write.csv(beta_sim, file = paste("beta_sim_", k, ".csv", sep = "")) write.csv(covar_sim, file = paste("covar_sim_", k, ".csv", sep = "")) write.csv(first_der, file = paste("first_der_", k, ".csv", sep = "")) write.csv(partial_der, file = paste("partial_der_", k, ".csv", sep = "")) } # Step 2: generate the necessory csv files for other TENET quantlets # number of columns in each partial derivative matrix cpd = 100 # number of rows in each partial derivative matrix rpd = (n - ws) library(miscTools) # since each firm does not regress on itself, we need to insert a zero column # vector in the position of every firms' partial derivative matrix vec_zero = matrix(0, rpd, 1) der.c = array(0, dim = c(rpd, cpd, cpd)) for (i in 1:100) { der.c[, , i] = insertCol(as.matrix(read.csv(file = paste("partial_der_", i, ".csv", sep = "")))[, 2:100], i, vec_zero) } # generate the connnectedness matrix con = array(0, dim = c(cpd, cpd, rpd)) for (i in 1:rpd) { con.v = rep(0, 100) for (j in 1:cpd) { con.v = rbind(con.v, der.c[i, , j]) } con[, , i] = con.v[-1, ] } # the date for the data files dt = as.Date(x0[, 1], format = "%d/%m/%Y")[(ws + 1):314] Date = strptime(as.character(dt), "%Y-%m-%d") Date1 = format(Date, "%d/%m/%Y") dt = as.data.frame(Date1) names(dt) = "Date" # the total connectedness total.c = rep(0, rpd) for (i in 1:rpd) { total.c[i] = sum(abs(con[, , i])) } # the average lambda series full.lambda = matrix(0, rpd, cpd) for (j in 1:100) { lambda.firm = read.csv(file = paste("lambda_sim_", j, ".csv", sep = "")) full.lambda[, j] = as.matrix(lambda.firm)[, 2] } average_lambda = 1/cpd * (rowSums(full.lambda)) tc_l = cbind(dt, total.c, average_lambda) # generate the necessory file for the quantlet TENET_total_connectedness write.csv(tc_l, file = "total_connectedness_and_averaged_lambda.csv", row.names = FALSE) # total in groups # in bank total.in.b = matrix(0, 266, 1) for (i in 1:rpd) { total.in.b[i] = sum(abs(con[, , i])[1:25, ]) } # in insurance total.in.ins = matrix(0, 266, 1) for (i in 1:rpd) { total.in.ins[i] = sum(abs(con[, , i])[26:50, ]) } # in broker dealer total.in.d = matrix(0, 266, 1) for (i in 1:rpd) { total.in.d[i] = sum(abs(con[, , i])[51:75, ]) } # in others total.in.o = matrix(0, 266, 1) for (i in 1:rpd) { total.in.o[i] = sum(abs(con[, , i])[76:100, ]) } tc_in = cbind(total.in.b, total.in.ins, total.in.d, total.in.o) colnames(tc_in) = c("Depositories_in", "Insurers_in", "Broker-Dealers_in", "Others_in") # total out groups # out bank total.out.b = matrix(0, 266, 1) for (i in 1:rpd) { total.out.b[i] = sum(abs(con[, , i])[, 1:25]) } # out insurance total.out.ins = matrix(0, 266, 1) for (i in 1:rpd) { total.out.ins[i] = sum(abs(con[, , i])[, 26:50]) } # out broker dealer total.out.d = matrix(0, 266, 1) for (i in 1:rpd) { total.out.d[i] = sum(abs(con[, , i])[, 51:75]) } # out others total.out.o = matrix(0, 266, 1) for (i in 1:rpd) { total.out.o[i] = sum(abs(con[, , i])[, 76:100]) } tc_out = cbind(total.out.b, total.out.ins, total.out.d, total.out.o) colnames(tc_out) = c("Depositories_out", "Insurers_out", "Broker-Dealers_out", "Others_out") tc_group = cbind(dt, tc_in, tc_out) # generate the necessory file for the quantlet TENET_total_in_out_groups write.csv(tc_group, file = "total_in_and_out.csv", row.names =