Bitcoin Price Forecasting (Time-Series)

Disclaimer

This is not financial advice. This is an applied technical exercise on time-series forecasting using historical Bitcoin closing price data.

Introduction

Given the recent turbulence of Bitcoin’s value in recent months, it is easy to succumb to hype and objection heresay from speculative sources. Often, many may lose sight of the simple fundamentals that lend credibility towards an asset’s valuation. With this in mind, the topic for today addresses a more technical method for speculating projected price.

Enter Rstudio’s Prophet package, developed by Facebook’s Core Data Science team. Prophet aims to streamline the forecasting process by implementing a regression model that fits non-linear trends with seasonality and holiday effects. This technique can be applied to daily, monthly, and yearly price prediction for such assets like crypto currency and securities.¹

Process

For our purposes, we will use Bitcoin as the asset in question to develop a one year forecast prediction for its closing price.

We will approach this simply and methodically by:

• Creating an Rproject directory to easily read in data.

• Downloading a dataset from Yahoo Finance on Bitcoin historical price data from September 2014 to June 2022.

• Isolating the Date and Close columns, then transform them to integrate into the Prophet package.

• Executing the fit model and forecast model to extrapolate Bitcoin’s closing price, one year from June 2022.

• Creating a vector that estimates the standard price forecast, upper bound price forecast, and lower bound price forecast.

• Condensing the new data into a mid-month interval for more efficient interpretation (This is not an essential step, but a personal preference).

• Viewing the final results in graphical and tabular form.

• Drawing a conclusion based off the forecasting estimates.

Step 1

Import libraries.

library(readxl) #to extract data from excel file
library(stringr) #to manipulate change string values for model integration 
library(dplyr) #to manipulate initial raw dataset
library(prophet) #to implement the times-series forecasting procedure

Step 2

From the Rproject directory, load the downloaded excel file into a vector; this is the preferred method to read in datasets for simplicity.²

my_data <- read_excel("BTC-USD.xlsx")

Step 3

Check the initial starting data and ending data to ensure the desired start time and end time, respectively.

knitr::kable(head(my_data))

knitr::kable(tail(my_data))

Step 4

Drop the Open, High, Low, Adj Close, and Volume columns; retain the Date & Close columns.

my_data2 = select(my_data, -2,-3,-4,-6,-7)
knitr::kable(head(my_data2))

Step 5

Replace Date column title with ds and Close column title with y. This is a crucial step in order for the Prophet package to read the data correctly.

colnames(my_data2)[1] <- "ds"
colnames(my_data2)[2] <- "y"
knitr::kable(head(my_data2))

Step 6

Replace all null string values with 0; this is crucial for the Prophet package to read the data correctly. Note: for this particular dataset the output is unchanged, but this is a precautionary and redundant step to ensure proper data cleaning.

my_data2$y <- str_replace_all(my_data2$y, 'null', '0') #replaces all nulls to 0s in column
knitr::kable(head(my_data2))

Step 7

Call the Prophet f(x) to Fit Model the model into a vector using daily seasonality…

Model1<-prophet(my_data2, daily.seasonality = TRUE)

…then, call the make_future_dataframe f(x) into another vector, to forecast 1 year from the dataset’s end date.

Future1<-make_future_dataframe(Model1, periods = (365*1))

Step 8

Create a vector that includes the Date ds, standard predicted close price yhat, lower estimated close price yhat_lower, and upper estimated close price yhat_upper.

Forecast1<-predict(Model1,Future1)
knitr::kable(tail(Forecast1[c('ds','yhat','yhat_lower','yhat_upper')]))

Step 9

Plot the Model Estimates.

dyplot.prophet(Model1,Forecast1)

This is the interactive graphical depiction of the forecasting model, in all its glory. The thin blue line is the forecasted price trend, while the dotted black line is the historical price over time.

prophet_plot_components(Model1,Forecast1)

These individual forecasted components show the seasonal trends captured at daily, weekly, and yearly intervals.

Step 10

Subset the dataframe for the 15th day of each month/year.

d<-data.frame(Forecast1[c('ds','yhat','yhat_lower','yhat_upper')])

Price_1_Year_Table<-d[format.Date(d$ds, "%d")=="15" &
                              !is.na(d$ds),]

Filter for the range between 2022-06-15 and 2023-06-15, as we are interested in this target year range for a grainier look.

Truncated <- (subset(Price_1_Year_Table, ds>="2022-05-15"))
knitr::kable(Truncated)

For extra credit, take the means of yhat, yhat lower, and yhat upper ranges.

knitr::kable(mean(Truncated$yhat))

The standard forecasted average for the next year.

knitr::kable(mean(Truncated$yhat_lower))

The lower bound forecasted average for the next year.

knitr::kable(mean(Truncated$yhat_upper))

The upper bound forecasted average for the next year.

Conclusion

Instinctively, the forecasted results indicate a sensible spread between projected Bitcoin closing price estimates, in the coming year. However, contrary to the package title, this is not an 100% future-predicting oracle analysis and should be considered with caution, given Bitcoin’s market volatility. Also, always consider black swan events.

Considering Bitcoin’s use case as an asset, this regression model bakes in some considerations as retail and institutional buying habits over time, factoring seasons and holidays, along with network difficulty (how difficult it is to mine a Bitcoin block, which offers staggered correlation to Bitcoin price)³.

However, this forecasting model does not necessarily account for the Federal Funds Rate changes with fidelity. A multiple linear regression analysis⁴ might be a better technique to pinpoint the significance in Federal Funds Rate dynamics on Bitcoin’s value.

Ultimately, this exercise demonstrates one tool in how one may project future seasonal trends in securities, commodities, or other non-monetary metrics such as: quantity of shoppers in a retail store, quantity of tickets sold, number of cars sold, or projected sale prices for merchandise. The key component to any analysis is accurate and well-scrubbed data entry.

For any questions or comments on this analysis please contact kris.kilgroe@sigmacode.io.

P.S.

Trust the math!

Session Info

sessionInfo()
#> R version 4.2.0 (2022-04-22)
#> Platform: x86_64-pc-linux-gnu (64-bit)
#> Running under: Linux Mint 20.3
#> 
#> Matrix products: default
#> BLAS:   /usr/lib/x86_64-linux-gnu/blas/libblas.so.3.9.0
#> LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.9.0
#> 
#> locale:
#>  [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
#>  [3] LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8    
#>  [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8   
#>  [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                 
#>  [9] LC_ADDRESS=C               LC_TELEPHONE=C            
#> [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       
#> 
#> attached base packages:
#> [1] stats     graphics  grDevices utils     datasets  methods   base     
#> 
#> other attached packages:
#> [1] prophet_1.0   rlang_1.0.2   Rcpp_1.0.8.3  dplyr_1.0.9   stringr_1.4.0
#> [6] readxl_1.4.0 
#> 
#> loaded via a namespace (and not attached):
#>  [1] lubridate_1.8.0      lattice_0.20-45      tidyr_1.2.0         
#>  [4] prettyunits_1.1.1    ps_1.7.1             zoo_1.8-10          
#>  [7] assertthat_0.2.1     digest_0.6.29        utf8_1.2.2          
#> [10] R6_2.5.1             cellranger_1.1.0     stats4_4.2.0        
#> [13] evaluate_0.15        ggplot2_3.3.6        highr_0.9           
#> [16] pillar_1.7.0         rstudioapi_0.13      callr_3.7.0         
#> [19] jquerylib_0.1.4      rmarkdown_2.14       labeling_0.4.2      
#> [22] htmlwidgets_1.5.4    loo_2.5.1            munsell_0.5.0       
#> [25] compiler_4.2.0       xfun_0.31            rstan_2.21.5        
#> [28] pkgconfig_2.0.3      pkgbuild_1.3.1       htmltools_0.5.2     
#> [31] tidyselect_1.1.2     tibble_3.1.7         gridExtra_2.3       
#> [34] bookdown_0.27        codetools_0.2-18     matrixStats_0.62.0  
#> [37] fansi_1.0.3          crayon_1.5.1         grid_4.2.0          
#> [40] jsonlite_1.8.0       gtable_0.3.0         lifecycle_1.0.1     
#> [43] DBI_1.1.3            magrittr_2.0.3       StanHeaders_2.21.0-7
#> [46] scales_1.2.0         rmdformats_1.0.4     RcppParallel_5.1.5  
#> [49] cli_3.3.0            stringi_1.7.6        farver_2.1.0        
#> [52] bslib_0.3.1          dygraphs_1.1.1.6     ellipsis_0.3.2      
#> [55] xts_0.12.1           generics_0.1.2       vctrs_0.4.1         
#> [58] tools_4.2.0          glue_1.6.2           purrr_0.3.4         
#> [61] processx_3.6.1       parallel_4.2.0       fastmap_1.1.0       
#> [64] yaml_2.3.5           inline_0.3.19        colorspace_2.0-3    
#> [67] extraDistr_1.9.1     knitr_1.39           sass_0.4.1

Footnotes