When adding de-listed data to the momentum system, performance degrades significantly, even over a short time horizon of one-month. I would expect to see performance degrade with longer minimum hold times such as six months or a year, but I am really surprised how much de-listed data affects even short-term holds.
I am re-running the last test of the momentum system with de-listed data and new code to account for stock splits, and while the test will not be done until late tonight, I can already see that performance will be degraded. I will post the results when they are ready so that we can observe how using unadjusted data affected the results vs. the adjusted data. (To see the impetus for the change of code and the thinking behind it, go here. Be sure to read the comments after the post.)
But the question remains: Why is de-listed data have such a profound affect on the system, even over short holds of 22 bars?
I am really curious to hear everyone’s thoughts on this issue.
Hi Woodshedder,
What kind of market cap and liquidity constraints are you imposing and how many stocks being held per month?
Okay, I read the original posts… Have you looked into your data to see how returns are calculated for stocks that are delisted. If they go to zero then 100% loss or 5% of your portfolio would be a pretty big hit. I would guess that a momentum stock that is probably near its 52-week highs (i mean its one of the top 20 stocks, right) that suddenly is “delisted” sounds to me much more likely to be a takeover or going private kind of thing than some sudden bankruptcy despite its strong price performance.
As far as the split-adjusted issue goes, you’re calculating momentum with the adjusted field but using minimum share price with the unadjusted field, right? Because if you’ve changed all calculations to unadjusted then you lose the “market cap momentum”.
Joshua, thanks for commenting. Yes, I’m calc’ing momo with the adjusted field but using the unadjusted for the filters.
I’ve been thinking the same as you, that buyouts and mergers were more likely than bankruptcy, since the stock would have been out-performing most of its peers.
My gut tells me your backtesting product is not handling the delisted instruments correctly. I’ve used delisted data from CSI for over 10 years on very long-term momentum systems and have not seen much (if any) of a degrade in performance.
Probably should generate a report of all your delisted instruments for a given year and then backtest over that year – analyze how your backtesting product marked those trades.
Hi Mike, my gut is telling me the same. Performance is just too awful with de-listed instruments added. I am going to guess that some of the trades are never being sold; AmiB is probably just holding them forever as it is probably not sure what to do with the stock after it is bought out/merged. The report is finished so I’ll look at the trade-by-trades this evening. Thanks!
Why don’t you try adding a conditional sell rule… if zero volume for 5 or so days in a row then sell. You’ll still have no way of knowing what your real returns would’ve been (assuming you’re not gonna go hunting bankruptcy and buyout data, lol) but hopefully any unlikely bankruptcies will be balanced out by takeovers. The problem of course is how many takeovers with their 10 to 30% premiums (guesstimate) relative to devastating losses from bankruptcies.
Josh
Not a bad idea, but the way AmiBroker handles rotational systems is that it will not even look at the position until the minimum bars held requirement is met. I’m going to have to query the user group and see what I can come up with. I’ve used logic before that says if this is the last bar, then close the position. That works fine with swing trading positions where the trade is evaluated daily. With the rotational logic it is different, and I don’t think the last bar, then sell, logic will work since it will first be waiting for the minimum bars held to be true.
Wood,
Since buyouts take some time to complete and frequently trap the stock in a narrow range between the announcement date and the completion/delist date, I suspect the issue is not (primarily) related to the handling of the delisting but rather to “illusory” momentum. What is probably happening is that the stock gaps way up (on the buyout news) triggering a high-momentum buy signal, then trades within a narrow range (less than 1%) for several weeks until the wide gap up is no longer included in the momentum calculation. At this point, the stock (which is may not yet be delisted) is sold by the system for lack of momentum, probably at a small gain or loss (1%, maybe 2%, in either direction from the initial buy point).
See the (lack of) action in XJT since the SkyWest acquisition bid on at the start of August as an example of what could be occurring in the backtest. As of today, any ROC value >= 14 shows an ROC of ~130+%, while any ROC < 14 is under 0.5%.
KTB
Good point KTB.
I think KTB has the right answer here. He is correctly applying Occam’s razor, IMO
I looked for this last night but did not immediately find any occurrences of this, although i was sure I would. I did not review but maybe 30-50 delisted issues trades. There are hundreds more, but I was still surprised not to find any instance of this.
Joshua has the idea. I use a stale condition with my simulator – if it has been 30 days since last price quote date…then position is closed.
Wood,
One more thing to account for, and you might already be on top of this, is Amibroker assigning negative values to large volume. I did not know anything about this, and it came up when I was looking over some data today. It probably would not be the cause of de-listed data expectations, but it could come up in adjusting the volume.
Actually, now that I think about it, this could be a huge problem adjusting the volume backwards, as the volume could be largely inflated for stocks that have had multiple splits. I actually found this on Citigroup, in late 2009.
http://www.premiumdata.net/support/amibroker.php#negativevolume
Correction: This has apparently been resolved in the most recent Amibroker version.
So much for staying up-to-date on my software.
lol…I hadn’t upgraded to the newest version yet…I was waiting for a good reason…this is enough!
I backed up my computer last night and upgraded my software, so I didn’t get much work done on this issue.
However, I did take about 20 minutes to review some of the trades on the de-listed issues. Not a single de-listed issue was bought out while it was held, and not a single issue was held open by the system longer than it should have been. Of course I did not get a chance to review the hundreds of de-listed issues, so this is anecdotal, at this point. Most of the de-listed issues were large losers. FWIW. I’ll take a more scientific approach in the future.
Found this paper on SSRN today…
“Survivorship Bias and Alternative Explanations of Momentum Effect”
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1663495
Haven’t read it so can’t comment but since you’re dealing with momo and delisted stocks it might be useful.
FWIW, I think long term momentum will be a struggling strategy for the next few years and with declining alpha beyond that. I don’t buy the EMH but I do think relative market efficiency increases over time and this would imply more mean reversion (albeit choppy, noisy mean reversion).
Josh
Thanks! Good find!
I would have originally guessed buy-outs per the above. Since that does not appear to be the major contributer consider this hypothesis: Stocks on the path to being delisted tend to be low priced stocks that can no longer meet minimum exchange requirements regarding price and fundamentals; and any small ($ wise) movement in cheap stocks can masquerade as high percentage momemtum – perhaps just a final pump before the final dump. To check this, consider running tests on different price brackets. My guess is that when you exclude delisted stocks, the low price stock bracket exhibits the very strong momentum performance, but when you include delisted stocks, the low priced stocks bracket shows significantly worse performance.
Actually, my version of Occam’s Razor would think that the under performance matches reality, fwiw.
But, I’ve been away from studying the market for most of the year.