Technology stocks have risen sharply from the second half of the previous year. Perhaps the rumors of the FED probable rate cuts affected that, but I want to look closer at it. I will examine this situation with a fund based mostly on NASDAQ-100 which is known heavily for technology stocks, and S&P Global 1200 Information Technology index.

library(tidyverse)
library(timetk)
library(tidyquant)

#İş Asset Technology Mixed Fund (ITP)
#(https://www.investing.com/etfs/tryispo00621)
df_itp <- read_csv("https://raw.githubusercontent.com/mesdi/blog/main/itp.csv")

#Tibble object
df_itp_tbl <- 
  df_itp %>% 
  mutate(Date = parse_date(Date, "%m/%d/%Y")) %>% 
  janitor::clean_names() %>% 
  select(date, itp = price) %>% 
  arrange(date)

#Federal Funds Effective Rate
df_fedfunds <- 
  tq_get("FEDFUNDS", get = "economic.data") %>% 
  #Low to high frequency
  pad_by_time(date, .by = "day") %>% 
  tidyr::fill(price, .direction = "up") %>% 
  select(date, fedfunds = price) %>% 
  #adding days to match the df_itp
  mutate(date = date + days(57)) 

#Merging all the data sets
df_merged <- 
  df_itp_tbl %>% 
  left_join(df_fedfunds)

I’ve added the FED effective rates to my model to improve the accuracy. Now, I will construct a causal inference model using a Bayesian time series structure. I will do that with CausalImpact package.

To do that, I set an intervention that probably affected the response variable (fund price). On March 30, 2023, the Federal Trade Commission (FTC) took action against several companies that used AI to create and distribute misleading advertisements. I will use this date as a pre and post-treatment period.

#Zoo object for the CausalImpact function
df_merged_zoo <- with(df_merged, zoo(cbind(itp, fedfunds), date))

#The pre-period and the post-period data sets
pre.period <- as.Date(c("2021-04-07", "2023-03-30"))
post.period <- as.Date(c("2023-03-31", "2024-05-28"))

#Causal effect
library(CausalImpact)

impact <- CausalImpact(df_merged_zoo, 
                       pre.period, 
                       post.period)

#Summarized results
summary(impact)

As seen in the above results, the relative effect shows that 66% increase which means the observed values are 66% higher than the counterfactual values which denotes how the response variable would have evolved if the intervention had never occurred. The result is significant because 95% confidence intervals(CI) exclude 0.

Finally, we will draw a plot showing the results we mentioned above.

#Plotting the effects
library(ggtext)

#Text box data frame for the plot
df_textbox <- data.frame(
  x = c(as.Date("2023-03-30"),
        as.Date("2023-11-01")),
  y = c(5,3),
  label = c("Actions against deceptive AI marketing",
            "Actions against unfair credit scoring algorithms"),
  family = "Bricolage Grotesque"
)

plot(impact, c("original")) +
  geom_vline(xintercept = as.Date("2023-11-01"),
             linetype = "dashed",
             color = "#a9a9a9",
             linewidth = 0.8)+
  geom_textbox(data = df_textbox, 
               aes(x = x, 
                   y = y, 
                   label = label,
                   family = family),
               width = unit(1.2, "inch"))+
  labs(title = "Increment of +66% in the ITP due to the actions taken by the FTC") +
  theme(plot.title = ggtext::element_markdown(face = "bold", 
                                              hjust = 0.5,
                                              size = 15),
        text = element_text(family = "Bricolage Grotesque"))