HBB announced earnings on Wednesday. Thursday the price fell somewhat. Friday the price fell some more. But today the price has fallen 20%.
What's the difference? FactSet didn't process the earnings until the weekend, and a ton of P123 users rebalance on Monday mornings.
I'm suggesting, though I can't be sure, that microcaps heavily favored by most users' ranking systems will be strongly affected by large changes in those rankings on Monday mornings.
This behavior is clearly self-defeating. No matter what your backtests may say, don't rebalance first thing Monday morning unless you're working with stocks that are unlikely to be affected by large orders placed by other P123 users. Backtests won't show you what other P123 users are doing and how that affects the price of the stocks you buy and sell. If they did, you'd probably be very much better off rebalancing on another day of the week.
I asked chatgpt what are Core Earnings for Q3 based on the sequential framework described here and got adjusted EPS at $0.79 (higher than PYQ).
Does it make any sense ? Thanks.
ChatGPT on the pension expenses:
The company incurred a significant one-time, non-cash expense of $7.6 million due to the termination of its U.S. defined benefit pension plan, which reclassified historical unrecognized losses from Accumulated Other Comprehensive Income into earnings. This expense is explicitly noted as nonrecurring
Indeed, FactSet, to its credit, lists $0.56 under SpcItemsPSQ, a large portion of which should be added to EPS. Strangely, Compustat does not, nor does its OPEPS take this into account. If you use the formula EPSExclXorQ+0.8*SpcItemsPSQ with FactSet, you get 0.59, which is closer to what chatgpt says. There's nothing in the income statement about the HealthBeacon Acquisition costs, though. Maybe ChatGPT picked that up from somewhere else. Both data providers have AcquisQ as 0.
Is it just P123 members having such a large impact or is it EVERYONE who uses FactSet data? How would we know from this one example?
In this example it was not just the large number of Monday trades done at P123 but also a coincidence in the FactSet data only becoming available over the weekend. What if the FactSet data for a different stock became available on a Tuesday? How would that effect this discussion?
I am not sure I am going to make a big change in my trading practices based on this one example.
The only clear takeaway from this, I believe, is that it would pay to upload earrings reports into ChatGPT on the day of the earnings release as Pitmaster alludes to in his post above. People might have sold Friday if ChatGPT has any predictive power at all. This post suggest that ChatGPT might indeed have some predictive power: Core Earnings LLM model
I wonder if an alert at rebalance that there is a recent earnings release for a ticker and even a link to the PDF of the earrings release for a member to upload into their favorite LLM would be a consideration, in addition to what has already been discussed.
These are great points, Jim. However, IIRC there is a lag between when when the earnings/filing is released and when the data becomes available in standardized form in the database.
The timing of the lag depends on the data service used
Someone looking at the earning/filings release (with or without ChatGPT), potentially could have avoided the drawdown on Monday. We are not saying the same thing?
ChatGPT may well be as bad at forecasting as a human being, but it is at least as good as a human being at reading financial reports and extracting financial metrics before they are available in the database.
To get back to the original subject of this post, HBB is trading far higher now than it was on Monday. So I would like to make the following suggestion. For those of you who place your trades manually, place them on Tuesday or Wednesday or Thursday. For those of you who place your trades automatically via IB, set a limit order on them. If any fail to execute, try again the next day or the day after. This will likely go contrary to your backtests, but your backtests cannot take into account what other P123 users are doing.
Personally, I'm very happy I did not sell HBB, which was my biggest holding in some of my strategies, on Monday, despite the huge drop in rank.
Focusing on the topic of which day to trade, I have some data in addition to what HBB did in this instance where Factset data was updated over the weekend (and had a one-time pension expense of $7.595 million dollars potentially misleading some algorithmic traders).
I have paper-traded ports that trade on specific days of the week at the close (close prices edited automatically by P123 with no transaction costs). Just one day of the week for each with the exception of holidays or in the case of my errors in missing a day and rebalancing the next day. Start date 12/1/23 for the port paper-traded first. So a little less than a year of data for each. !5 stocks for each port. Basically the Easy To Trade Universe (minimal changes with minimal increases in liquidity).
While the current price action aligns with expectations, it may not fully capture the underlying dynamics. HBB’s beta, around 0.66, suggests that the stock typically moves less aggressively than the market as a whole. In this case, the sharp decline on specific news was idiosyncratic to HBB, unrelated to broader market trends. On other days with no specific news, its performance aligned more closely with the general market’s direction.
This approach – considering both theoretical and practical factors when analyzing stock returns – highlights a nuanced view on beta’s role in explaining HBB’s behavior. It’s something I’ve advocated for over time, and I believe Portfolio123 could benefit from integrating this perspective, given the supporting empirical evidence
The two largest increases from EXFY and RIGL both had significant atypical gains starting on Friday 11/08. Evidently there were others ahead of the P123 investors.