Here are the three questions I think this dataset answers best. Each one rests on the profile I ran across all 422,729 rows, which cover 152 tickers from 2012-01-02 to 2025-08-04. I haven't yet run the analyses themselves. **1. Did the AI surge change price levels and trading activity, and for which companies?** - About 99,984 rows (23.7%) fall after the ChatGPT cutoff used in my queries. That is enough history on both sides to compare before and after. - The comparison would use each company's growth in close price and its change in volume against its own pre-2023 baseline. - Ranking companies by that change would show whether the gains were broad or concentrated in a few AI-exposed names. That matters for overweight and underweight decisions. - The gold tables `semi_conductor_se_by_company_name_date__yyyy` and `..._yyyy_mm` already hold these aggregates. **2. Which stocks carry the most risk, and did volatility rise after the surge?** - The average daily high-low range is 3.69% of price, but that is a pooled figure. Risk probably differs a lot between large, established chipmakers and small, thinly traded names. - I would compute the average range per company and compare it before and after the surge. Volatility may have moved differently from price, for example rising in the most AI-exposed names. - This would help size positions and pick companies for a sector basket. **3. When does volume signal unusual interest, and can it be separated from data gaps?** - Volume is extremely skewed. The median is 898,500 shares a day, the mean is about 8.47M, and the maximum is about 3.69B. - That means a few huge spikes dominate the averages. Volume should be judged relative to each stock's own history, for instance as a rolling z-score or a multiple of its median. - 4,888 rows have zero volume. These are probably halted days, missing data, or illiquid names, and they must be separated from genuine quiet days before any spike analysis. - I would check whether spikes cluster around specific dates or move together across companies. Clustering would point to sector-wide events rather than stock-specific ones. **Caveat:** the dataset has no fundamentals or exchange field, only price and volume. Any "AI exposure" grouping would have to come from outside knowledge of the companies, not from the data. The card lists about 149 companies while my profile shows 152, so I would check for duplicate or renamed tickers before ranking companies.