See my PM, please do not reply to this poster. Please try to ask your own questions, and if you're interested in seeing my plan for attacking such liquidity issues, let's just say I'm hatching a deal to obtain significant capital asset funding for my new mutual fund. <b>Quote of the Day?</b? Are you opposed to greenmail, dude? <i>Would you take a check from me?</i> Hopefully, only people playing me in Ice Hockey will ever know what that really means. Peace.-- Beau Wolinsky
In spite of my many valid points, and the work I did to actually improve the so called Pairs System (which actually isn't pairs trading at all), BoWo is choosing to ignore me. The reason is obvious... he doesn't know what he's doing and doesn't even understand the issues I have raised. Shame... this thread could have been useful and people could have benefitted. However, in the interest of everyone's education, let me show you one of several reasons why BoWo's QID QLD backtest is completely invalid. Attached find a spreadsheet that covers the full history of the QID. It includes settlement prices and 1600 prices. BoWo's backtest was done to settlement prices, but you can see that at times this price is VERY different from the price you could have executed. Now... you have another quirky instrument on the other side of the "pair" (in quotes because this is most definitely NOT a pairs methodology... see BoWo's own Collective2 admission I posted a few weeks ago) so the problem is magnified. Systems like this are extremely sensitive to small variations in price. BoWo himself posts prices to 8 decimal places (which is extremely amusing, but whatev.) Your backtest was done on fantasy prices. This is a problem with ALL pairs backtests on daily data, and explains why BoWo thinks daily pairs systems are better than intraday. He is testing on fantasy data and backtests on this fantasy data do in fact look better... but they are prices you cannot possibly execute with any reliability. On one hand, this isn't BoWos fault. If you don't really understand markets and spend your entire day watching markets, you would not realize this. In short, if you aren't a trader (he isn't) you don't see issues like this unless someone points them out to you. However, the problem is that BoWo sets himself up as an expert... his lack of knowledge on this topic alone should prove he is not. (This is not the first time he has shown his total lack of experience in posts here on ET.... interesting reading in that thread.) BoWo... there are about 6 posts of mine in this thread that demand your attention... I raised valid points and did a ton of work for you. You owe it to me and to your readers to respond. It is your obligation which you assumed by creating this thread and posting your system calls. I posted logical critiques and raised serious questions... how can you, in good conscience, ignore them? bs2167: It is not, as BoWo says, a liquidity issue at all. Absolutely not. It's a misunderstanding that strikes right at the heart of the system concept... much more serious than a liquidity issue. In short, BoWo missed something very serious.
This also assumes Bwol has a quality data source. Yahoo OHLC on the Q's is one thing, but, even then - assuming you're sampling the actual close @ 1600 (not the settlement), are you really going to get that price? Aside from liquidity at close, the conservative thing to do (i.e. for modeling consistency) is to sample the 1555 bar or the 1559 bar close/volume to make a proper fill assessment. Then again, this is basic stuff... Just for kicks, this is from a thread I participated in a few months ago where I introduced a (pseudo? hehe ) random system based on the 12:50pm - 12:55pm (PST) coin-flip. http://www.elitetrader.com/vb/showthread.php?s=&threadid=179402&perpage=10&pagenumber=42 The core of this system is a little effect that can be amplified significantly based on some nice position sizing and a - get this - a random number generator! The code and result is attached (I used TS because its quick). What does this mean? Well - lets try this experiment, just for kicks: 1. For the original BoWo, lets flip a coin when any signal is presented, i.e. ANY long, short, sell or cover signal. I know Wealth Lab has a Rand() function of some kind. 2. Go long if heads or go short if tails no matter what the original signal was. 3. Run it a few times and lets see some results. The major point here is, I can take an effect - add some random conditions and produce a very very good backtest! Next step is to get major funding for this novel system! Haha.:eek: Mike
Talon, Have been following this thread as I am in the process of creating a pairs based strategy myself. A lot of insight here for anyone following this thread. And for those who are, I strongly reccommend you read up Talon's other thread he has going if you havent already. Talon, if you dont mind could you elaborate a little more on the volatilty based filter you added? are we talking something about as simple as a short term average range/Standard Deviation being higher/lower than a longer term average range/ Standard Deviation?
Wow - really surprised at the magnitude of the differences. For the unitiated like myself, here is some info on how the 16:00 vs. settlement discrepancy occurs: http://www.sec.gov/answers/closepr.htm In an earlier instance, Mike was nice enough to alert me to the dangers of using a daily data source. And to his point, I've yet to find an instance where a daily data provider does not use the settlement price as the daily close. Obviously this has major implications for constructing a realistic database. However, there is still one thing I am unclear on regarding the ability to reliably transact at the settlement price. According to the paper below, it appears that market-on-close or transactable limit-on-close orders have/will always trade at the CTA's reported settlement price. Of course this ignores the possibility that your order would have moved the market. Is the potential for moving the market the only factor that makes using the settlement price unreliable - or is there something else I am missing? http://www.sec.gov/rules/sro/nyse/2009/34-59345.pdf
You retail traders have no idea what you can do. I plan to be on the CME pit when I go through with this very soon. You can facebook me by sending a message.
Bowo, How about taking talon off ignore (again) and answering his last question? Also, you might want to take a look at his thread to find out what a real trading system template looks like. This ought to be one step towards knowing how to even recognize something tangible. Mike
Copy of publicly available pairs trading black box. This is straight out of WL4 developer downloads. You could run it yourself. 96% of equity straight out of QID and QLD black box no mods. B&H shown. Long + Short Buy & Hold $100,000.00 $100,000.00 $152,516.36 $30,819.47 $52,516.36 ($69,180.53) 52.52% -69.18% 12.95% -28.80% 20.75% 100.23% 210 1 $250.08 ($69,180.53) 1.08% -69.07% 4.28 871 138 0 65.71% 0.00% $153,328.61 $0.00 $1,111.08 $0.00 4.21% 0.00% 3.15 0 18 N/A 72 1 34.29% 100.00% ($100,812.25) ($69,180.53) ($1,400.17) ($69,180.53) -4.93% -69.07% 6.43 871 10 N/A ($26,496.80) ($108,817.58) -15.01% -77.93% 11/17/2009 12/29/2009 53.04 -51.12 1.52 0 1.98 0.64 0.85 0 1.03 -0.44 5.11 44.38 4.14 17.79 3.13 -1.62 3.98 0 1.34 0 0.24 -0.24 Annually Period Starting $ Return % Return % Max DD Exposure Entries Exits 7/13/2006 7,811.36 7.81 -1.80 13.98 22 20 1/3/2007 7,799.88 7.23 -7.94 20.91 68 68 1/2/2008 43,264.08 37.42 -14.64 18.15 60 62 1/2/2009 -6,358.95 -4.00 -15.01 24.84 60 58 Fuck, I lagged by a bit than just the blackbox, perhaps I should let my parameters last a lot longer?-- B Position Symbol Shares Entry Date Entry Price Exit Date Exit Price % Change Long QID 772,946 3/11/2009 60.72 3/12/2009 57.61 -5.13 Long QLD 1,730,059 3/31/2009 26.95 4/7/2009 29.08 7.9 Long QLD 1,711,024 4/8/2009 29.4 4/14/2009 30.7 4.44 Long QLD 1,702,112 4/21/2009 29.95 4/22/2009 30.59 2.13 Long QLD 1,529,838 5/12/2009 34.42 5/13/2009 32.83 -4.62 Long QLD 1,586,727 5/14/2009 31.76 5/19/2009 33.73 6.2 Long QID 1,643,788 6/2/2009 32.71 6/4/2009 32.32 -1.19 Long QID 1,582,579 6/16/2009 32.85 6/17/2009 33.41 1.71 Long QLD 1,498,938 6/23/2009 35.77 6/25/2009 36.11 0.95 Long QLD 1,565,507 7/8/2009 34.71 7/10/2009 34.55 -0.46 Long QLD 1,369,975 7/16/2009 38.79 7/28/2009 44.96 15.9 Long QLD 1,362,437 8/6/2009 45.43 8/13/2009 46.07 1.41 Long QLD 1,390,140 8/17/2009 43.55 8/21/2009 45.78 5.12 Long QLD 1,486,059 9/2/2009 43.68 9/4/2009 44.74 2.43 Long QID 2,790,622 9/11/2009 23.94 9/14/2009 24.28 1.42 Long QLD 1,342,964 9/25/2009 49.88 9/29/2009 50.74 1.73 Long QLD 1,355,081 9/30/2009 51.33 10/1/2009 48.7 -5.12 Long QLD 1,366,785 10/2/2009 47.34 10/7/2009 50.04 5.7 Long QLD 1,312,861 10/27/2009 52.54 10/28/2009 50.12 -4.6 Long QLD 1,355,111 10/29/2009 49.42 10/30/2009 49.98 1.14 Long QLD 1,401,137 11/2/2009 47.81 11/4/2009 48.94 2.36 Long QID 3,503,546 12/23/2009 19.43 Open Open -3.09 Estimated sustainability: 2225*500*10,000=$11,125,000,000.00 When we get there, I will be maxed.