r-只在特定日期进行滚动回归

Question

我正在尝试进行以下滚动回归：stock_return=α+βmarket_return。

我的数据集"数据"如下：

Company Date        stock_return    market_return       Alpha   Beta
AAPL    01.01.2014  3%              4%
…
AAPL    31.12.2019  5%              1%
MSFT    01.01.2014  2%              4%
…
MSFT    31.12.2019  6%              1%

此外，我有一个数据帧，其中包含应该执行回归的事件：

Company Date        
AAPL    05.02.2015
…   
MSFT    04.08.2018

我正在使用滚动回归的runner包：

running_regression <- function(z) {
coef(lm(stock_return ~ market_return , data = as.data.frame(z)))
}
output <- runner(seq_along(data$market_return), 
k = 180, 
lag = 5,
at = ?
f = running_regression)

然而，为了对第二个数据帧中的事件进行回归，我很难处理我的函数以及将什么放入"at"中。

请参阅下面的一个最小示例：

Date <- seq(from = as.Date("2014-01-01"), to = as.Date("2019-12-31"), by = 'day')
Date <- format(Date, format="%d.%m.%Y")
Company.name <- c(replicate(2191, ""))
AAPL <- data.frame(Company.name = "AAPL",Date)
market_return <- c(rnorm(2191))
AAPL <- cbind(AAPL, market_return)
MSFT <- c(replicate(2191, "MSFT"))
MSFT <- data.frame(Company.name = "MSFT",Date)
MSFT <- cbind(MSFT, market_return)
df <- rbind(AAPL, MSFT)
stock_return <- c(rnorm(4382))
df <- cbind(df,stock_return)
df[,"Alpha"] <- NA
df[,"Beta"] <- NA
Company.name2 <- c(replicate(450, "AAPL"),replicate(450, "MSFT"))
Event_date <- sample(seq(as.Date('2015/01/01'), as.Date('2019/12/31'), by="day"), 900)
Event_date <- format(Event_date, format="%d.%m.%Y")
df2 <- data.frame(Company.name2, Event_date)
#____
running_regression <- function(z) {
coef(lmList(stock_return ~ market_return | Company.name, data=z))
}

runner(df$market_return, 
k = 180, 
lag = 5,
f = running_regression)

Answer 1

由于运行程序版本0.3.5，您可以指定x = df并从data.frame在窗口上运行回归。因为你突变了df2，你必须用x = df[df$Company.name == Company.name2,]将df的相关部分子集化——你必须对idx做同样的事情。

running_regression_intercept <- function(x) {
coef(lm(stock_return ~ market_return, data = x))[1]
}
running_regression_slope <- function(x) {
coef(lm(stock_return ~ market_return, data = x))[2]
}

library(dplyr)
library(runner)
df2 %>%
group_by(Company.name2) %>%
mutate(
intercept = runner(
x = df[df$Company.name ==  Company.name2[1], ],
k = "180 days",
lag = "5 days",
idx = df$Date[df$Company.name == Company.name2[1]],
at = Event_date,
f = running_regression_intercept,
),
slope = runner(
x = df[df$Company.name == Company.name2[1], ],
k = "180 days",
lag = "5 days",
idx = df$Date[df$Company.name == Company.name2[1]],
at = Event_date,
f = running_regression_slope
)
)
# Company.name2 Event_date    alpha     beta
#     <fct>         <date>        <dbl>    <dbl>
#   1 AAPL          2017-01-12  0.0114   0.00488
#   2 AAPL          2017-07-31 -0.0654   0.00574
#   3 AAPL          2019-02-27 -0.0861   0.0310 
#   4 AAPL          2018-09-06  0.0405  -0.0630 
#   5 AAPL          2015-09-03 -0.121   -0.0246 
#   6 AAPL          2018-11-20 -0.0283  -0.0254 
#   7 AAPL          2015-07-03 -0.116   -0.0186 
#   8 AAPL          2015-02-03  0.102    0.0409 
#   9 AAPL          2017-03-16 -0.0157   0.0124 
#   10 AAPL          2019-06-08 -0.00302  0.0532 

---

我需要修改您的数据，因为格式已将Event.Date从Date更改为character。

Date <- seq(from = as.Date("2014-01-01"), 
to = as.Date("2019-12-31"), 
by = 'day')
market_return <- c(rnorm(2191))
AAPL <- data.frame(
Company.name = "AAPL", 
Date =  Date,
market_return = market_return
)
MSFT <- data.frame(
Company.name = "MSFT", 
Date = Date,
market_return = market_return
)
df <- rbind(AAPL, MSFT)
df$stock_return <- c(rnorm(4382))
df <- df[order(df$Date),]
df2 <- data.frame(
Company.name2 = c(replicate(450, "AAPL"), replicate(450, "MSFT")), 
Event_date = sample(
seq(as.Date('2015/01/01'), 
as.Date('2019/12/31'), 
by="day"),
size =  900)
)

Answer 2

希望我答对了，

如果您只想在df2中的特定日期运行回归，一个快速的解决方案是使用相关名称合并它们：

head(merge(df2,df,by.x=c("Company.name2","Event_date"),by.y=c("Company.name","Date")))
Company.name2 Event_date market_return stock_return Alpha Beta
1          AAPL 01.01.2016    -0.4814525   -0.9864268    NA   NA
2          AAPL 01.01.2019    -1.3093155    0.8357707    NA   NA
3          AAPL 01.02.2017     1.5059626   -1.3181592    NA   NA
4          AAPL 01.03.2015    -1.3226590    1.4786993    NA   NA
5          AAPL 01.03.2016     2.1394338   -0.8900286    NA   NA
6          AAPL 01.03.2019     0.6035526    0.5453212    NA   NA

你可以看到日期搞砸了，所以先按公司和日期对你的df2进行排序是有意义的：

df2 = df2[order(df2$Company.name2,
as.Date(as.character(df2$Event_date),format="%d.%m.%Y")),]
df_merged = merge(df2,df,
by.x=c("Company.name2","Event_date"),
by.y=c("Company.name","Date"),sort=FALSE)

要进行滚动回归，您可以在这里使用lmList，但它会变得有点复杂。下面我使用roll_lm from roll来计算系数。。不太了解你的滞后，也许你可以详细说明，但你不可能对跑步者进行回归：

library(roll)
result = by(df_merged,df_merged$Company.name2,function(z){
roll_lm(z$market_return,z$stock_return,180)})
tail(result$AAPL$coefficients)
(Intercept)         x1
[445,] -0.07817682 0.10662762
[446,] -0.06440454 0.09257577
[447,] -0.07007445 0.09461642
[448,] -0.05917523 0.09582312
[449,] -0.05292590 0.10025369
[450,] -0.04930798 0.09911921