在R Tidyquant中更有效地找到股市支持趋势线

Question

我已经开发了一些R代码来查找和绘制股票市场数据的趋势线。但是，我使用的方法涉及对处理能力的暴力使用，并且可能需要很长时间，特别是如果我想要绘制超过一年的价格数据的趋势线。所以，如果有人能帮我找到一种更有效的方法，我会很高兴。

基本上，我目前的方法包括在数据集中生成所有可能的两个日低对，生成通过一对点的所有可能的趋势线，然后测试每条线，看看数据集中是否有日低跌破这条线。我们保留所有为FALSE的行。

随着您尝试生成趋势线的时间范围的增加，所需的处理时间呈指数级增加。为了减少处理时间，我一直在过滤低于简单移动平均线的低点数据集。这去除了大约一半的数据，通常保留了最相关的数据点。然而，这并不能完全解决问题。当分析多个自动收报机符号上的长时间帧时，运行此代码仍然需要很长时间。

下面是我得到的信息：

# Load libraries for tidy stock data analysis
library(tidyverse)
library(tidyquant)

# Retrieve 1 year's worth of Apple stock price data from Yahoo! Finance
ticker <- "AAPL"
start <- Sys.Date() %m-% years(1)
prices <- tq_get(ticker, from = start) %>%
  mutate(open = round(open,digits=2),
         high = round(high,digits=2),
         low = round(low,digits=2),
         close = round(close,digits=2)) %>%
  select(symbol,date,open,high,low,close)

# Filter prices data for lows that are below the simple moving average
lows <- prices %>%
  filter(low < SMA(close),date<max(date))

# Find all unique possible combinations of two lows
# (and all unique possible combinations of their associated dates)
all_lowcombos <- bind_cols(as.data.frame(t(combn(lows$date,m=2,simplify=TRUE))),as.data.frame(t(combn(lows$low,m=2,simplify=TRUE))))
colnames(all_lowcombos) <- c("X1","X2","Y1","Y2")

# Generate a trendline for every combination of points
n <- seq(1:nrow(all_lowcombos))
low_trendfinder <- function(n,all_lowcombos){
  model <- lm(c(all_lowcombos$Y1[n],all_lowcombos$Y2[n])~c(all_lowcombos$X1[n],all_lowcombos$X2[n]))
  data.frame(intercept = model$coefficients[1],slope = model$coefficients[2])
}
low_trendlines <- map_dfr(n,low_trendfinder,all_lowcombos = all_lowcombos)

  # For each low_trendline, check if any low in the prices dataframe falls below the line
  # Keep only trendlines for which this is FALSE
  # Also make sure the trendline wouldn't be less than half the current price for today's date; I only want lines that might be tradeable in the next week
  low_trendline_test <- function(x,y,prices){
    !any(x*as.numeric(prices$date) + y > prices$low + 0.01) & !(x*as.numeric(Sys.Date())+y < 0.5*prices$close[nrow(prices)])
  }
  none_below <- map2(.x = low_trendlines$slope,.y = low_trendlines$intercept,.f = low_trendline_test,prices = prices)
  none_below <- unlist(none_below)
  low_trendlines <- low_trendlines[none_below,]

# Chart support trendlines on a candlestick chart
prices %>% ggplot(aes(x = date, y = close)) + 
  geom_candlestick(aes(open = open, high = high, low = low, close = close)) + 
  geom_abline(intercept=low_trendlines$intercept,slope=low_trendlines$slope) + 
  labs(title = paste(ticker,"Trendline Chart"), 
       y = "Price", 
       x = "Date", 
       caption = paste("Price data courtesy of Yahoo! Finance. Accessed ",
                       Sys.Date(),
                       ".",
                       sep="")) + 
  theme_tq()

​

​

Answer 1

以下是来自评论的解决方案的实现。

# Load libraries for tidy stock data analysis
library(tidyverse)
library(tidyquant)

# Retrieve 1 year's worth of Apple stock price data from Yahoo! Finance
ticker <- "AAPL"
start <- Sys.Date() %m-% years(2)
prices <- tq_get(ticker, from = start) %>%
  mutate(open = round(open,digits=2),
         high = round(high,digits=2),
         low = round(low,digits=2),
         close = round(close,digits=2)) %>%
  select(symbol,date,open,high,low,close)

# Filter prices data for lows that fall on the convex hull
lows <- prices[chull(prices[c("date", "low")]),] %>%
  filter(date<max(date))

# Find all unique possible combinations of two lows
# (and all unique possible combinations of their associated dates)
all_lowcombos <- bind_cols(as.data.frame(t(combn(lows$date,m=2,simplify=TRUE))),as.data.frame(t(combn(lows$low,m=2,simplify=TRUE))))
colnames(all_lowcombos) <- c("X1","X2","Y1","Y2")

# Generate a trend line for every combination of points
n <- seq(1:nrow(all_lowcombos))
low_trendfinder <- function(n,all_lowcombos){
  model <- lm(c(all_lowcombos$Y1[n],all_lowcombos$Y2[n])~c(all_lowcombos$X1[n],all_lowcombos$X2[n]))
  data.frame(intercept = model$coefficients[1],slope = model$coefficients[2])
}
low_trendlines <- map_dfr(n,low_trendfinder,all_lowcombos = all_lowcombos)

# For each low_trendline, check if any low in the prices dataframe falls below the line
# Keep only trendlines for which this is FALSE
# Also make sure the trendline wouldn't be less than half the current price for today's date
low_trendline_test <- function(x,y,prices){
  !any(x*as.numeric(prices$date) + y > prices$low + 0.01) & !(x*as.numeric(Sys.Date())+y < 0.5*prices$close[nrow(prices)])
}
none_below <- map2(.x = low_trendlines$slope,.y = low_trendlines$intercept,.f = low_trendline_test,prices = prices)
none_below <- unlist(none_below)
low_trendlines <- low_trendlines[none_below,]

# Chart support and resistance trendlines and this week's price targets
prices %>%
  ggplot(aes(x = date, y = close)) +
  geom_candlestick(aes(open = open, high = high, low = low, close = close)) +
  geom_abline(intercept=low_trendlines$intercept,slope=low_trendlines$slope) +
  labs(title = paste(ticker,"Trendline Chart"), y = "Price", x = "Date", caption = paste("Price data courtesy of Yahoo! Finance. Accessed ",Sys.Date(),".",sep="")) +
  theme_tq()

​

​