Here are three questions this dataset can answer well. The figures below come from the profiling queries I ran. The questions themselves are proposals, and I haven't run the analyses behind them yet. ## What the data covers - There are 422,729 daily rows for 152 companies and tickers, from 2012-01-02 to 2025-08-04. Each row has open, high, low, close and volume. - 99,984 rows (23.7%) fall after the ChatGPT launch, so there is a large pre-AI baseline to compare against. - The average daily high-low range is about 3.69% of price. - Volume is heavily skewed. The median is 898,500 shares, the mean is about 8.47M, and the maximum is about 3.69 billion. - 4,888 rows have zero volume. ## 1. Did the AI surge change the stocks, and did it lift all of them or only a few? Compare each company's price return, volatility (range %) and volume before and after late 2022. Then test whether the gains are concentrated in a handful of names or spread across the sector. The data holds about 10 years of history before the surge and about 2.7 years after it. That supports a per-company before/after comparison, and a ranking of which companies changed most. - **Decision it serves:** which stocks to overweight or underweight. - **Method:** a `company_name × year` aggregation. The gold tables `semi_conductor_se_by_company_name_date__yyyy` and `..._yyyy_mm` already provide this. ## 2. Has risk changed, and which companies carry the most of it? Track the daily range (high−low)/close by company and over time. Look at how volatility clusters around events such as 2020, 2022 and 2025. Then check whether higher volatility comes with higher returns, or whether some names are simply riskier. - **Decision it serves:** position sizing and sector basket construction. - **Method:** a scatter of return against average range % per company, plus a time series of sector-wide range %. ## 3. When does volume signal unusual interest, and does it lead price moves? Volume has a median of 898.5K against a mean of 8.47M, so typical days and spike days are very different. Flag days where volume is far above a company's own rolling norm. Then test whether price moves over the following days differ after those spikes, and whether spikes have become more common since 2023. - **Decision it serves:** timing entries and exits. - **Method:** per-company rolling volume baselines, using `raw.semi_conductor_se`. Exclude the 4,888 zero-volume rows, or treat them as a data-quality issue. They may mark halted days, missing data, or gaps in early history for some tickers. ## Caveats for any of these - Prices are not confirmed as split- or dividend-adjusted, so long-horizon returns need checking before you trust them. - The card's "~149 companies" differs slightly from the 152 I counted. - The dataset is observational, so results show association rather than proof that AI caused a change.