trading_strategies.Rmd

---
title: "Trading-strategies"
author: "Brent Morrison"
date: "`r Sys.Date()`"
output:
  html_document:
    fig_caption: yes
    theme: spacelab 
    highlight: pygments
    toc: TRUE
    toc_depth: 3
    number_sections: FALSE
    code_folding: hide
    toc_float:
      smooth_scroll: FALSE
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE, warning = FALSE, message = FALSE, error = FALSE)
```

Situation

https://www.statestreet.com/web/insights/articles/documents/state-street-ssa-event-time-full-report-050422_4728159.GBL.pdf

Complication

Question

Answer

```{r}
# Libraries
library(dplyr)
library(tidyr)
library(slider)
library(DBI)
library(RPostgres)
library(lubridate)
library(ggplot2)
library(cowplot)
library(tidyquant)


# Set default theme
def_theme1 <- theme_minimal() +
  theme(
    legend.title = element_blank(),
    legend.position = c(0.9,0.9),
    legend.background = element_blank(),
    legend.key = element_blank(),
    plot.caption = element_text(size = 8, color = "grey55", face = 'italic'), 
    axis.title.y = element_text(size = 8, color = "darkslategrey"),
    axis.title.x = element_text(size = 8, color = "darkslategrey"),
    axis.text.y = element_text(size = 7, color = "darkslategrey"),
    axis.text.x = element_text(size = 7, color = "darkslategrey")
    )


# Custom functions
perc_range <- function(x){
  current <- tail(x, 1)
  min <- min(x)
  range <- max(x) - min
  res <- ( current - min ) / range
  return(res)
}

perc_pos<- function(x){
  x <- x[!is.na(x)]
  res <- sum(x > 0) / length(x)
  return(res)
}

rtn_from_high <- function(x){
  current <- tail(x, 1)
  max <- max(x)
  res <- log( current / max )
  return(res)
}


# Database connection
config <- jsonlite::read_json('C:/Users/brent/Documents/VS_Code/postgres/postgres/config.json')


con <- dbConnect(
  RPostgres::Postgres(),
  host      = 'localhost',
  port      = '5432',
  dbname    = 'stock_master',
  user      = 'postgres',
  password  = config$pg_password
)
```

## The data

Text here

```{r}

# Stock data
qry_test <- "select * from access_layer.return_attributes where date_stamp > current_date - interval '20 years' order by 1, 2"
qry_send <- dbSendQuery(conn = con, statement = qry_test) 
df_raw <- dbFetch(qry_send)

# S&P500 data
sp5_sql <- "select * from access_layer.daily_sp500_ts_vw"
sp5_send <- dbSendQuery(conn = con, statement = sp5_sql) 
sp5_raw_df <- dbFetch(sp5_send)

vix <- tq_get("VIXCLS", get = "economic.data")
```


```{r}
pr12 <-  c('0.75','0.80','0.5','0.4')   # perc_range_12m
pp12 <-  c('0.75','0.70','0.5','0.4')   # perc_pos_12m
ra12s <- c('0.05','0.07','0.2','0.0')   # rtn_ari_12m_sct
ra12i <- c('0.05','0.07','0.2','0.0')   # rtn_ari_12m_ind

r1 <- sprintf("perc_range_12m > %s & perc_pos_12m > %s & rtn_ari_12m_sct > %s & rtn_ari_12m_ind > %s", pr12[1], pp12[1], ra12s[1], ra12i[1])
r2 <- sprintf("perc_range_12m > %s & perc_pos_12m > %s & rtn_ari_12m_sct > %s & rtn_ari_12m_ind > %s", pr12[2], pp12[2], ra12s[2], ra12i[2])
r3 <- sprintf("perc_range_12m < %s & perc_pos_12m < %s & rtn_ari_12m_sct < %s & rtn_ari_12m_ind < %s", pr12[3], pp12[3], ra12s[3], ra12i[3])
r4 <- sprintf("perc_range_12m < %s & perc_pos_12m < %s & rtn_ari_12m_sct < %s & rtn_ari_12m_ind < %s", pr12[4], pp12[4], ra12s[4], ra12i[4])
```



```{r}
# Trading strategy
df1 <- df_raw %>% 
  group_by(symbol) %>%
  #filter(symbol %in% c('A','AA')) %>% 
  mutate(
    fwd_rtn_1m = lead((adjusted_close-lag(adjusted_close))/lag(adjusted_close)),
    perc_range_12m = slide_dbl(.x = adjusted_close, .f = perc_range, .before = 11, .complete = TRUE),
    perc_pos_12m = slide_dbl(.x = rtn_log_1m, .f = perc_pos, .before = 11, .complete = TRUE),
    rtn_from_high_12m = slide_dbl(.x = adjusted_close, .f = rtn_from_high, .before = 11, .complete = TRUE)
  ) %>% 
  ungroup() %>% 
  group_by(sector) %>%
  mutate(
    rtn_ari_3m_sct = mean(rtn_ari_3m),
    rtn_ari_12m_sct = mean(rtn_ari_12m)
  ) %>% 
  ungroup() %>% 
  group_by(industry) %>%
  mutate(
    rtn_ari_3m_ind = mean(rtn_ari_3m),
    rtn_ari_12m_ind = mean(rtn_ari_12m)
  ) %>% 
  ungroup() %>%
  mutate(
    ind1 = if_else(eval(parse(text = r1)), 1, NaN), # use NaN for failing to meet trading rule, 0 disrupts mean
    ind2 = if_else(eval(parse(text = r2)), 1, NaN),
    ind3 = if_else(eval(parse(text = r3)), -1, NaN),
    ind4 = if_else(eval(parse(text = r4)), -1, NaN),
    fwd_rtn1 = fwd_rtn_1m * ind1,
    fwd_rtn2 = fwd_rtn_1m * ind2,
    fwd_rtn3 = fwd_rtn_1m * ind3,
    fwd_rtn4 = fwd_rtn_1m * ind4
  ) %>%
  select(
    symbol, date_stamp, adjusted_close, 
    ind1, ind2, ind3, ind4, 
    fwd_rtn1, fwd_rtn2, fwd_rtn3, fwd_rtn4,
    fwd_rtn_1m, rtn_log_1m, rtn_ari_3m_sct, rtn_ari_12m_sct, rtn_ari_3m_ind, 
    rtn_ari_12m_ind, perc_range_12m, perc_pos_12m, rtn_from_high_12m
  ) 


# Aggregate trading rule outcome to portfolio
df2 <- df1 %>% 
  group_by(date_stamp) %>% 
  summarise(
    strat_rtn_lag1 = mean(fwd_rtn1, na.rm = TRUE), n1 = sum(!is.na(fwd_rtn1)),
    strat_rtn_lag2 = mean(fwd_rtn2, na.rm = TRUE), n2 = sum(!is.na(fwd_rtn2)),
    strat_rtn_lag3 = mean(fwd_rtn3, na.rm = TRUE), n3 = sum(!is.na(fwd_rtn3)),
    strat_rtn_lag4 = mean(fwd_rtn4, na.rm = TRUE), n4 = sum(!is.na(fwd_rtn4))
  ) %>% 
  mutate(
    strat_rtn1 = lag(strat_rtn_lag1, n = 1, default = 0),
    strat_rtn2 = lag(strat_rtn_lag2, n = 1, default = 0),
    strat_rtn3 = lag(strat_rtn_lag3, n = 1, default = 0),
    strat_rtn4 = lag(strat_rtn_lag4, n = 1, default = 0)
  ) %>% 
  filter(date_stamp >= as.Date('2014-10-31')) %>%
  replace_na(list(
    strat_rtn_lag1 = 0, strat_rtn_lag2 = 0, strat_rtn_lag3 = 0, strat_rtn_lag4 = 0, 
    strat_rtn1 = 0, strat_rtn2 = 0, strat_rtn3 = 0, strat_rtn4 = 0
  )) %>% 
  mutate(
    strat_value1 = cumprod(1 + strat_rtn1),
    strat_value2 = cumprod(1 + strat_rtn2),
    strat_value3 = cumprod(1 + strat_rtn3),
    strat_value4 = cumprod(1 + strat_rtn4)
  )


# Stack data
df3 <- df2 %>% 
  select(date_stamp, strat_value1, strat_value2, strat_value3, strat_value4) %>% 
  pivot_longer(
    cols = c(strat_value1, strat_value2, strat_value3, strat_value4),
    names_to = 'label', 
    values_to = 'strat_value'
  ) %>% 
  arrange(label, date_stamp)


# Prepare S&P500 data for plotting
sp5_monthly <- sp5_raw_df %>% 
  group_by(date_stamp = floor_date(date_stamp, "month")) %>% 
  mutate(date_stamp = ceiling_date(date_stamp, unit = "month") - 1) %>% 
  summarise(
    adjusted_close = last(adjusted_close),
    volume = mean(volume)
  ) %>% 
  ungroup() %>% 
  filter(date_stamp >= as.Date(min(df2$date_stamp)))

sp5_monthly$strat_value <- sp5_monthly$adjusted_close / sp5_monthly$adjusted_close[1]
sp5_monthly$label <- 'S&P 500 index'


# Stack data
data <- bind_rows(
  sp5_monthly[c('date_stamp', 'label','strat_value')], 
  df3[c('date_stamp', 'label','strat_value')]
  )
```

```{r}
 p1 <- ggplot(data, aes(x = date_stamp, y = strat_value, linetype = label)) + 
  geom_line() +
  geom_text(data = data[data$date_stamp == max(data$date_stamp),], 
            aes(label = label, x = date_stamp + 365/4,  y = strat_value), size = 2.5) +
  #facet_wrap(vars(label)) +
  labs(x = '',
       y = 'Strategy returns',
       title = 'S&P 500 and custom momentum strategy returns',
       subtitle = 'Strategy formation rules described below',
       caption = "Source: IEX Cloud and Alpha Vantage via 'Stock Master' database") +
  scale_y_continuous(breaks = seq(-2,4, 0.5)) +
  scale_x_date(date_breaks = '1 years',
               date_labels = '%Y') + 
  def_theme1 + 
  #theme(legend.position = c(0.1, 0.8)) +
  theme(legend.position = 'none')

p1
```

```{r}
vix_monthly <- vix %>% 
  fill(price) %>% 
  group_by(date_stamp = floor_date(date, "month")) %>% 
  mutate(date_stamp = ceiling_date(date_stamp, unit = "month") - 1) %>% 
  summarise(close = last(price))
```

```{r}
p2 <-  filter(vix_monthly, date_stamp >= as.Date('2014-10-31'), date_stamp <= max(data$date_stamp)) %>%
  ggplot(aes(x = date_stamp, y = close)) + 
  geom_line() +
  geom_text(data = vix_monthly[vix_monthly$date_stamp == max(data$date_stamp),], 
            aes(label = close, x = date_stamp + 365/4), size = 2.5) +
  labs(x = '',
       y = 'VIX Index',
       title = '',
       subtitle = '',
       caption = "Source: Nasdaq / Quandl") +
  scale_y_continuous(breaks = seq(0, 60, 15)) +
  scale_x_date(date_breaks = '1 years',
               date_labels = '%Y') + 
  def_theme1 + 
  theme(legend.position = 'none')

p2
```

```{r}
plot_grid(p1, p2, ncol = 1, align = 'v', axis = 'lr')
```

```{r}
library(patchwork)
p1 + p2 + plot_layout(ncol = 1)
```

```{r}
dbDisconnect(con)
```