# Three analytical questions this dataset can answer The dataset holds daily open, high, low, close and volume for 152 semiconductor stocks, from 2012-01-02 to 2025-08-04 (422,729 rows). Each row is one stock on one trading day. About 99,984 rows (23.7%) fall in the post-ChatGPT period, so there is enough data on both sides of the AI surge to compare before and after. ## 1. Did the AI surge change which semiconductor stocks lead, and by how much? - **What to compute:** Compare each company's price growth before and after late 2022, using the yearly and company-level gold tables. Look at the ranking of winners and laggards, not just the average. - **Why it matters:** This supports overweight and underweight decisions. The median close across all rows is only about $35.57, so a sector average would hide large differences between a few AI leaders and the rest. - **Caveat:** Prices are raw closes, so stock splits may distort long-run growth comparisons. ## 2. Has risk changed, with volatility measured as the daily high-low range? - **What to compute:** Compare each company's average daily range (high minus low, as a share of price) before and after the AI surge. - **Baseline:** The average daily range is about 3.69% across the full history. - **Why it matters:** It shows whether the AI era made the sector riskier per holding, and which names carry the most risk for position sizing. - **Extension:** Add a breakdown by year to see whether volatility is trending or concentrated in particular shocks. ## 3. When does trading volume signal unusual interest, and is that interest broad or concentrated? - **What to compute:** Flag days when a stock's volume far exceeds its own typical level, then check whether those days cluster after the AI surge or in a few companies. - **Baseline:** Mean daily volume is about 8.47M shares but the median is only 898,500. The maximum is about 3.69 billion shares, so volume is very heavily skewed. - **Why it matters:** Comparing each stock to its own history, rather than to a sector average, gives a usable entry and exit signal. - **Data quality:** 4,888 rows have zero volume. Treat them as gaps or illiquid days and exclude or flag them first. ## Limits - There are no fundamentals, such as earnings or market cap, so the data shows what happened to prices and volume, not why. - The dataset is observational, so any "AI effect" is a before/after association rather than a causal result. - The profile of the data I ran shows 152 distinct companies, slightly more than the roughly 149 on the dataset card.