Financial Data Science: Stock Market Trend Analysis

Published with the Cornell Data Journal alongside co-authors Cole Gaines and Karen Petrosyan, this project investigates the "January Effect": the long-contested hypothesis that stocks post stronger returns in January as investors repurchase shares they sold off in November and December for tax-loss harvesting.

Rather than take the hypothesis on faith, the project tests it directly against 20+ years of industry-wide return data pulled from the Yahoo Finance API, using two technical indicators: a seasonality index of monthly returns, and the Relative Strength Index (RSI), which signals whether a security is being overbought or oversold in a given month.

A seasonality index built across five major indices (S&P 500, Dow Jones, NASDAQ, Russell, and sector funds) from 2010–2023 confirms the pattern: January, July, and November show the highest excess returns of the year, lining up with both the January Effect and the mid-year index-rebalancing bump.

Comparing the distribution of January returns against every other month makes the gap concrete: January's mean return sits well above the rest of the year, even though day-to-day volatility still overlaps.

Seasonality index of mean returns by month across five indices, 2010–2023.

Distribution of January returns vs. every other month.

Small-cap stocks are more volatile than large-caps, so one hypothesis was that tax-loss selling would hit them harder heading into January and set up a sharper rebound. Breaking excess January returns out by market-cap tier shows no clean trend, suggesting the seasonal effect operates market-wide or sector-by-sector rather than by company size.

Excess January returns broken out by market-cap tier, from micro-cap to mega-cap.

20-year average monthly RSI values, computed per ticker and rolled up into industry heatmaps, tell a more sector-specific story. Technology stocks spike in RSI around May–July, not January, while airlines and entertainment names both run hot from November through February.

Rolling every industry into one market-wide heatmap shows automobiles and retail peaking hardest in November (RSI of 60.7 and 61.5), with most sectors elevated from November through February, consistent with the January Effect window but not uniform across the board.

20-year average monthly RSI heatmap for 8 major technology tickers.

Market-wide RSI heatmap comparing 8 industries month by month.

The data supports the existence of the January Effect and related seasonal tendencies, but not as a market-wide rule: some industries show clear seasonal buying and selling pressure while others don't, and the effect doesn't track cleanly with company size. Seasonality looks less like a single market-wide signal and more like something that plays out differently sector by sector.