Financial and economic analysis and data visualization
By Christopher C. Smith
INTRODUCTION
The modelingmarkets repo contains financial data analysis work by Christopher Carroll Smith. Chris writes about quantitative finance on Substack at https://modelingmarkets.substack.com/. He also podcasts about finance on the Wall Street Petting Zoo podcast, available through your favorite podcasting app. Code is created in R and Python. Sample outputs are included in the repo. Code is free to use with proper acknowledgement of authorship under terms of the Apache 2.0 license.
SCRIPTS FOUND IN THIS REPOSITORY
10yr-yield-forecaster.Rmd - Uses multiple regression models trained with five-fold cross-validation to forecast the 10-year US Treasury yield and S&P 500 P/E ratio from date and PCE inflation rate.
C19SectorReturns.R - Charts stock market returns by equal weight S&P 500 sector since the beginning of the Covid-19 pandemic using data from Yahoo Finance.
c19-return-comparer/app.R - Reactive version of my C19SectorReturns.R script, created with Plotly and Shiny. (Directory includes both the app and the data, which data was fetched using the C19SectorReturns.R script above.) See the app live on the web here: https://chriscarrollsmith.shinyapps.io/c19-return-comparer/.
DownFriday-DownMonday.R - Tests Jeffrey Hirsch's "Down Friday + Down Monday" indicator to see if DF+DM day pairs are more commonly associated with market inflection points than other day-pairs. Uses split-adjusted $SPY price data from Yahoo Finance and tests for robustness against 24 different model specifications. Effect size is positive in all model specifications. Effect size is larger on shorter time frames (5-10 days) and appears weaker when analysis is extended to 15-20 days. Measured magnitude of excess inflection is <1% of index price in all specifications. "Excess inflection" associated with the DF+DM signal is proportionally largest immediately after the signal appears. 3 days after the signal, magnitude of inflection is 18-43% larger than the magnitude of inflection of the average non-DF+DM day-pair. It hits a minimum 18 days after the signal, when magnitude of inflection is only 0.5-13% larger than the magnitude of inflection of the average non-DF+DM day pair.
historicalvolume.R - Pulls stock price and historical volume data from Yahoo Finance via the tidyquant library and constructs a profile of historical volume. Inspired by TradingView's "volume profile," this charting technique seeks to identify price levels where a lot of historical volume was located. The theory is that if a price had a lot of historical volume, then there will still be buy and sell orders located there that could make it a critical turning point for the stock's price action.
multi-asset-model.R - Constructs simple forecasting models for several different US and global stock and bond index ETFs. Uses supervised machine learning to select model inputs, and multiple linear regression to forecast returns. Sharpe ratio is calculated based on forecasted returns and observed standard deviation from model when forecasting out of sample on a test set. (Sharpe ratios must be used with caution because error is systematically biased rather than random.)
stockmarketseasonality.R - Script to determine the seasonal pattern of returns for any ticker over the last 20 years (or for the lifetime of the ticker, if shorter than that). Just change the ticker and type variables and execute the script to get the chart you want.
trendlinefinder.R - Scrapes Yahoo! Finance data and charts it using tidyquant functions. Finds and draws all major support and resistance trend lines.
tsadata.R - Scrapes TSA traveler throughput data from tsa.gov and charts 2020 throughput against 2019 throughput to examine the impact of the Covid-19 (coronavirus, SARS-COV-2) pandemic.