Prediction in R with Prophet
23/10/2024 Yufy Firdiansyah, MM, S.Sos
Short Introduction
In the ever-evolving landscape of financial markets, accurately forecasting price movements is crucial for making informed investments. This analysis demonstrates how to leverage the power of the Prophet package in R, designed for forecasting time series data. Prophet simplifies the complexities associated with time series modeling, allowing users to capture trends, seasonality, and holiday effects effectively. By using historical price data, we will explore how Prophet can deliver robust forecasts, enabling investors and analysts to make better decisions based on anticipated market conditions. This introduction provides a foundation for understanding the capabilities of Prophet and sets the stage for practical implementations in financial forecasting.
Install Package & Loading Libraries
#Loading Libraries - Please install if necessary
library(readr)
library(dplyr)
library(stringr)
library(quantmod)
library(TTR)
library(ggplot2)
library(prophet)
# Clear the environment to avoid naming conflicts
rm(list = ls())Data Cleaning
Data cleaning is essential because downloads from Investing.com typically lack a Close Price column for candlestick charts, require renaming the ‘Vol.’ column to ‘Volume’, and have Date, Vol., and Change columns in character format.
First we set up the directory and upload the file to be analyzed
#Set Directory and Upload The File
setwd("~/RProject/Stocks_2")
stocs <-read.csv("LPKR Historical Data.csv")
head(stocs)## Date Price Open High Low Vol. Change..
## 1 10/01/2024 105 100 106 100 244.35M 3.96%
## 2 09/30/2024 101 100 101 98 179.18M 2.02%
## 3 09/27/2024 99 100 101 96 170.50M 0.00%
## 4 09/26/2024 99 100 101 98 150.01M -1.00%
## 5 09/25/2024 100 103 104 98 237.06M -1.96%
## 6 09/24/2024 102 99 105 97 417.81M 4.08%Then we trim column names to tidy up the data
column_names <- colnames(stocs)
print(column_names)## [1] "Date" "Price" "Open" "High" "Low" "Vol." "Change.."str(stocs)## 'data.frame': 176 obs. of 7 variables:
## $ Date : chr "10/01/2024" "09/30/2024" "09/27/2024" "09/26/2024" ...
## $ Price : int 105 101 99 99 100 102 98 101 107 95 ...
## $ Open : int 100 100 100 100 103 99 101 107 114 88 ...
## $ High : int 106 101 101 101 104 105 103 108 116 98 ...
## $ Low : int 100 98 96 98 98 97 95 101 98 88 ...
## $ Vol. : chr "244.35M" "179.18M" "170.50M" "150.01M" ...
## $ Change..: chr "3.96%" "2.02%" "0.00%" "-1.00%" ...colnames(stocs) <- trimws(colnames(stocs))
head(stocs)## Date Price Open High Low Vol. Change..
## 1 10/01/2024 105 100 106 100 244.35M 3.96%
## 2 09/30/2024 101 100 101 98 179.18M 2.02%
## 3 09/27/2024 99 100 101 96 170.50M 0.00%
## 4 09/26/2024 99 100 101 98 150.01M -1.00%
## 5 09/25/2024 100 103 104 98 237.06M -1.96%
## 6 09/24/2024 102 99 105 97 417.81M 4.08%Do not miss the Date Column because it is still in character/string format
#Convert Date to Date Format
library(lubridate)
stocs$Date <- parse_date_time(stocs$Date, orders = c("ymd", "dmy", "mdy"), quiet = TRUE)Next, we create close column from open column next row
# Create the 'Close' column with the 'Open' value from the next row
socs <- stocs %>%
mutate(Close = lead(Open))
# Print the updated dataframe
head(socs)## Date Price Open High Low Vol. Change.. Close
## 1 2024-10-01 105 100 106 100 244.35M 3.96% 100
## 2 2024-09-30 101 100 101 98 179.18M 2.02% 100
## 3 2024-09-27 99 100 101 96 170.50M 0.00% 100
## 4 2024-09-26 99 100 101 98 150.01M -1.00% 103
## 5 2024-09-25 100 103 104 98 237.06M -1.96% 99
## 6 2024-09-24 102 99 105 97 417.81M 4.08% 101We also have to delete the latest row because it does not have value inside close cell
#Delete latest row
socs_data_cleaned <- na.omit(socs)
str(socs_data_cleaned)## 'data.frame': 175 obs. of 8 variables:
## $ Date : POSIXct, format: "2024-10-01" "2024-09-30" ...
## $ Price : int 105 101 99 99 100 102 98 101 107 95 ...
## $ Open : int 100 100 100 100 103 99 101 107 114 88 ...
## $ High : int 106 101 101 101 104 105 103 108 116 98 ...
## $ Low : int 100 98 96 98 98 97 95 101 98 88 ...
## $ Vol. : chr "244.35M" "179.18M" "170.50M" "150.01M" ...
## $ Change..: chr "3.96%" "2.02%" "0.00%" "-1.00%" ...
## $ Close : int 100 100 100 103 99 101 107 114 88 87 ...
## - attr(*, "na.action")= 'omit' Named int 176
## ..- attr(*, "names")= chr "176"Predictions
First we prepare
# Prepare the data for Prophet, ensuring columns are named correctly
df_prophet <- data.frame(
ds = socs_data_cleaned$Date, # Date column
y = socs_data_cleaned$Close # Close price column
)
head(df_prophet)## ds y
## 1 2024-10-01 100
## 2 2024-09-30 100
## 3 2024-09-27 100
## 4 2024-09-26 103
## 5 2024-09-25 99
## 6 2024-09-24 101Then we create the model
# Fit the model
model <- prophet(df_prophet)
# View the model summary
summary(model)## Length Class Mode
## growth 1 -none- character
## changepoints 25 POSIXct numeric
## n.changepoints 1 -none- numeric
## changepoint.range 1 -none- numeric
## yearly.seasonality 1 -none- character
## weekly.seasonality 1 -none- character
## daily.seasonality 1 -none- character
## holidays 0 -none- NULL
## seasonality.mode 1 -none- character
## seasonality.prior.scale 1 -none- numeric
## changepoint.prior.scale 1 -none- numeric
## holidays.prior.scale 1 -none- numeric
## mcmc.samples 1 -none- numeric
## interval.width 1 -none- numeric
## uncertainty.samples 1 -none- numeric
## specified.changepoints 1 -none- logical
## start 1 POSIXct numeric
## y.scale 1 -none- numeric
## logistic.floor 1 -none- logical
## t.scale 1 -none- numeric
## changepoints.t 25 -none- numeric
## seasonalities 1 -none- list
## extra_regressors 0 -none- list
## country_holidays 0 -none- NULL
## stan.fit 4 -none- list
## params 6 -none- list
## history 5 data.frame list
## history.dates 175 POSIXct numeric
## train.holiday.names 0 -none- NULL
## train.component.cols 3 data.frame list
## component.modes 2 -none- list
## fit.kwargs 0 -none- listNow, Predict 60 days
# Create a dataframe for future dates (e.g., predicting the next 60 days)
future <- make_future_dataframe(model, periods = 60)
# Make predictions
forecast <- predict(model, future)
# View the forecast results
head(forecast)## ds trend additive_terms additive_terms_lower additive_terms_upper
## 1 2024-01-03 89.16507 -0.6878961 -0.6878961 -0.6878961
## 2 2024-01-04 88.99320 -2.4165668 -2.4165668 -2.4165668
## 3 2024-01-05 88.82132 -2.2542357 -2.2542357 -2.2542357
## 4 2024-01-08 88.30569 -0.9600463 -0.9600463 -0.9600463
## 5 2024-01-09 88.13382 -1.3154280 -1.3154280 -1.3154280
## 6 2024-01-10 87.96194 -0.6878961 -0.6878961 -0.6878961
## weekly weekly_lower weekly_upper multiplicative_terms
## 1 -0.6878961 -0.6878961 -0.6878961 0
## 2 -2.4165668 -2.4165668 -2.4165668 0
## 3 -2.2542357 -2.2542357 -2.2542357 0
## 4 -0.9600463 -0.9600463 -0.9600463 0
## 5 -1.3154280 -1.3154280 -1.3154280 0
## 6 -0.6878961 -0.6878961 -0.6878961 0
## multiplicative_terms_lower multiplicative_terms_upper yhat_lower yhat_upper
## 1 0 0 79.76347 97.04436
## 2 0 0 78.10065 95.76970
## 3 0 0 77.83975 94.70307
## 4 0 0 78.40013 95.51043
## 5 0 0 78.29185 95.41208
## 6 0 0 78.79586 96.02870
## trend_lower trend_upper yhat
## 1 89.16507 89.16507 88.47718
## 2 88.99320 88.99320 86.57663
## 3 88.82132 88.82132 86.56708
## 4 88.30569 88.30569 87.34565
## 5 88.13382 88.13382 86.81839
## 6 87.96194 87.96194 87.27405Then we visualize
# Plot the forecast
plot(model, forecast)
# Optional: Plot components to analyze seasonality and trend
prophet_plot_components(model, forecast)
Conclusion
In conclusion, our prediction indicates an upward trend in prices. Additionally, we observed notable weekly patterns and a consistent monthly trend, highlighting the cyclical nature of price movements. These insights not only reinforce the efficacy of the Prophet model in capturing these dynamics but also provide valuable information for making informed investment decisions