# Three analytical questions this dataset could answer I ran out of steps before running any of the analyses below. What I did establish is the dataset's shape and scale, and the three questions follow from that. None of them has been answered yet. ## What the data contains (verified) - **Scope:** 422,729 daily rows covering 152 semiconductor-related companies, from 2012-01-02 to 2025-08-04. There is one ticker per company. - **Columns:** open, high, low, close and volume, plus company and date. - **Completeness:** close and volume have no nulls. There are 4,888 zero-volume rows, about 1.2% of the total. - **Post-ChatGPT window:** 99,984 rows (23.7%) fall on or after the ChatGPT launch, and 96,711 rows are from 2023 onward. - **Typical values:** - Median close is about $35.57. - Average daily range is about 3.69% of close. - Median daily volume is 898,500, against a mean of about 8.47M. The max is about 3.69B, so volume is heavily right-skewed. - **Pre-aggregated gold tables:** these exist by company, by year, by month and by day, which makes time-series and cross-sectional comparisons cheap to run. ## The three questions ### 1. Did the AI boom (post-ChatGPT) change volatility and trading activity across semiconductor stocks? - **Why it fits:** the data spans a long pre-ChatGPT baseline and about 99,984 post-ChatGPT rows. - **What to compare:** average daily range % of close, and median volume (not mean, given the skew), before and after launch. - **Follow-up:** whether the change is broad or concentrated in a few AI-exposed names. - **Status:** not tested. I only have the overall average range (3.69%), with no pre/post split. ### 2. Which companies drove returns and how unevenly? Did a few winners dominate the post-2022 period? - **Why it fits:** `semi_conductor_se_by_company_name_date__yyyy` has 1,785 company-year rows of open and close, so yearly returns per company can be calculated. - **What to compare:** the spread of returns across the 152 companies, and whether the top few account for most of the gains after 2022 versus earlier years. - **Status:** not tested. ### 3. How do semiconductor stocks move together, and did that correlation change over time? - **Why it fits:** the company-by-month table has 20,499 rows, enough to build a return matrix. - **What to compare:** average pairwise correlation by year, for example whether it rose during stress periods or diverged after 2023. - **Follow-up:** whether there are identifiable clusters, such as large caps versus small caps. - **Status:** not tested. ## Caveats to address before any analysis - **Zero-volume rows (4,888):** these could be halted or illiquid days, or data gaps. They should be excluded or examined. - **Volume outliers:** the 3.69B maximum is far above the median, so check it before using volume averages. - **Uneven histories:** the 1,785 company-year rows are fewer than 152 × 14 = 2,128. Some companies therefore have shorter histories (for example, later IPOs), which could bias cross-company comparisons. - **Price adjustments:** it is unconfirmed whether prices are split- or dividend-adjusted. This matters for any return calculation. - **Range metric:** the average range % appeared in the tool output as both 0.0369 and 3.69. These are the same figure expressed as a fraction and as a percent.