The paper identifies its dataset as the Center for Research and Security Prices, which is often abbreviated as CRSP and can be referred to as CRISP. This is a fairly well-known pricing dataset. And if you check out their website, you can see that the data is available via subscription. We don’t use that data precisely, but we are using a pricing dataset which is very similar. It’s also helpful to check what stock universe the authors think are most relevant to the factors they are analyzing. Even though a factor would ideally be applicable and useful for a broad range of stocks, it’s also possible that certain factors only generate meaningful signals for a certain subgroup of stocks. In the abstract, the authors mentioned that mean reversion or momentum factors that they’re evaluating are more pronounced for hard to value firms. If we look through the paper for mention of harder to value firms, we see how they define them in section four titled, Short-term Overnight Return Persistence, Hard to Value Firms and Institutional Share Holdings. According to other papers they cite, harder to value firms are more volatile, have a smaller market cap, are newer companies, currently less profitable, and considered high growth companies as exhibited by higher price to earnings ratios. Remember that I said when you see and, think conditional factor. If you wanted to implement these specific ideas, you would do so via the mechanics of a conditional factor. We don’t do that in this exercise however. The authors provide some context for why harder to value firms may be more likely to exhibit overbuying at market open or weekly momentum of closed open returns. The hypothesis is that individual investors may rely more on sentiment when the fundamentals of the company are more difficult to measure. This sentiment is expressed in the recent closed open returns of the stock. We can also scan the paper for useful methodology, which we can apply more broadly to any factors that we’re evaluating. For instance, in section two, the authors define the weekly overnight return on the stock. If you haven’t read this section yet, you may imagine taking a window of five days daily returns, adding them up to get the weekly cumulative close to open return. The authors do something slightly different. They take the average of the daily returns over five trading days, then they multiply by five. Can you think of how this is different or why they may do this? Notice that if we had missing data, then simply adding up the daily closed open returns over a five-day window, may not make each week comparable. If a certain week is missing data for one day, then we would only have four days worth of returns to add up. By taking the daily average and then multiplying by five, we can better deal with missing data and help make each week’s cumulative return more comparable to other weeks.