muck Posted April 1, 2011 Author Share Posted April 1, 2011 I for one am following it carefully. Thanks for making it available to us. +1 Thanks. Quote Link to comment Share on other sites More sharing options...
Brentastic Posted April 1, 2011 Share Posted April 1, 2011 Another "+1" week. ...anyone still looking at this, or am I wasting my time? I'm following Quote Link to comment Share on other sites More sharing options...
Ursa Majoris Posted April 1, 2011 Share Posted April 1, 2011 Thanks. Even us communists take care of our portfolios. Quote Link to comment Share on other sites More sharing options...
ljbrun Posted April 1, 2011 Share Posted April 1, 2011 Still following. Thank you very much. Quote Link to comment Share on other sites More sharing options...
matt770 Posted April 2, 2011 Share Posted April 2, 2011 Still following. Thank you very much. +1 Quote Link to comment Share on other sites More sharing options...
muck Posted April 15, 2011 Author Share Posted April 15, 2011 Sorry ... missed last week .... Last week was another "+1" and this week remains a "+1". ********************************************************** It is still a 50/50 proposition about the prospective client I mentioned to you a month or so ago. It looks like that if he's going to invest, I'll be tweaking the model a bit, with one of the possible side effects being that there will be a bit finer gradation between signal strength (i.e., -1.0 ... -0.5 ... 0.0 ... +0.5 ... +1.0 ... +1.5 ... +2.0 ... +2.5 ...). I will keep you guys updated along the way... Thanks! Quote Link to comment Share on other sites More sharing options...
Ursa Majoris Posted April 15, 2011 Share Posted April 15, 2011 Still hanging in there with your model. Quote Link to comment Share on other sites More sharing options...
muck Posted April 16, 2011 Author Share Posted April 16, 2011 Per the request of my-hoped-for-new-client, I think I'm nearly finished with the final tweaking of some of the rules and methodology ... all of which is intended to give additional 'color' to the expectations for the coming week(s). The new 'feature' is a "cash" signal, in addition to new "strong sell" and "really strong buy" signals. While I will need to re-vet the formulas (to make sure everything is referencing what it should be), it looks like the following outputs will be used: -2.25 ----- strong sell ----- 16 weeks of 1401 in the period back to June 1, 1984 ----- average weekly return -2.87% -1.50 ----- sell / short ----- 122 weeks of 1401 in the period ----- average weekly return -0.77% 0.00 ----- cash ----- 508 weeks of 1401 in the period ----- average weekly return -0.03% 0.75 ----- buy / long ----- 183 weeks of 1401 in the period ----- average weekly return +0.14% 1.50 ----- strong buy ----- 547 weeks of 1401 in the period ----- average weekly return +0.61% 2.25 ----- really strong buy ----- 25 weeks in 1401 in the period ----- average weekly return of +1.24% The "score" reflects the leverage used when trading ... so a score of "0.75" would result in holding $75 for every $100 of cash invested in the account ... and a score of "-1.50" would result in being short $150 for every $100 of cash invested in the account ... etc ... The leverage assumptions are reflected in the return calculations below. *********************************************** Again, assuming that the formulas are as they should be, the new signals for the calendar year (and the next weeks' performance) would have been as follows: Week ending 12/31/10 = 0.00 (market was up +1.10% the next week) Week ending 1/7/11 = 0.00 (market was up +1.71% the next week) Week ending 1/14/11 = 0.00 (market was down -0.76% the next week) Week ending 1/21/11 = 1.50 (market was down -0.55% the next week) Week ending 1/28/11 = 1.50 (market was up +2.71% the next week) Week ending 2/4/11 = 0.00 (market was up +1.39% the next week) Week ending 2/11/11 = 0.00 (market was up +1.04% the next week) Week ending 2/18/11 = 0.00 (market was down -1.72% the next week) Week ending 2/25/11 = 1.50 (market was up +0.15% the next week) Week ending 3/4/11 = 1.50 (market was down -1.28% the next week) Week ending 3/11/11 = 1.50 (market was down -1.92% the next week) Week ending 3/18/11 = 1.50 (market was up +2.70% the next week) Week ending 3/25/11 = 0.00 (market was up +1.42% the next week) Week ending 4/1/11 = 0.00 (market was down -0.32% the next week) Week ending 4/8/11 = 1.50 (market was down -1.22% the next week) Week ending 4/15/11 = 0.75 Year to Date = +0.6% Last 52 Weeks = +14.7% Last 3 Years = +43.9% (annualized) Last 5 Years = +30.6% (annualized) Last 10 Years = +41.8% (annualized) Last 15 Years = +42.2% (annualized) *********************************************** For the time being, I will report both scores until I finalize any revisions to the methodology. *********************************************** NOTE: During the crash of 1987, the revised model was showing a score of 1.50 (i.e., it lost money on leverage) During the crash of Aug/Oct 2008, the revised model had an average score of -1.75 (i.e., it was short on leverage and made a killing) ...no systematic model is perfect... ...caveat emptor... Quote Link to comment Share on other sites More sharing options...
Ursa Majoris Posted April 16, 2011 Share Posted April 16, 2011 Muck, I hope you have a backup of this thing on USB or external hard drive or something. Interesting that until this week the only signals were 0.00 and 1.50 despite the market fluctuating up and down the following week. FWIW, I still think this year is a good time to be holding but keep both eyes open for the unexpected. Biggest thing to look for might be the debt ceiling fight, which will go all the way to the end, methinks. Quote Link to comment Share on other sites More sharing options...
muck Posted April 22, 2011 Author Share Posted April 22, 2011 Original Model = +1.00 (i.e., buy) Revised Model = 0.00 (i.e., cash) NOTE: In the revised model, the signal indicates to be in cash approximately 35% of the time, so the trading is quite a bit more frequent ... so, don't think that this is some sort of a precursor to "something really bad" (i.e., "go short because the sky is falling"). Quote Link to comment Share on other sites More sharing options...
muck Posted April 22, 2011 Author Share Posted April 22, 2011 Muck, I hope you have a backup of this thing on USB or external hard drive or something. Interesting that until this week the only signals were 0.00 and 1.50 despite the market fluctuating up and down the following week. FWIW, I still think this year is a good time to be holding but keep both eyes open for the unexpected. Biggest thing to look for might be the debt ceiling fight, which will go all the way to the end, methinks. Trading with perfect information would result in quite extraordinary returns. It's not that big of a deal to miss a move (or even to be wrong) in the short term --- all I'm doing is trying to shift the odds to being right, and then when the odds are tilted in my favor to increase my bets a bit. Maybe its a bit like trying to count cards while playing black-jack? Quote Link to comment Share on other sites More sharing options...
Brentastic Posted April 22, 2011 Share Posted April 22, 2011 Good stuff, Muck Quote Link to comment Share on other sites More sharing options...
muck Posted April 26, 2011 Author Share Posted April 26, 2011 ...he said the revised model was too volatile...so, back to the drawing board...and out comes another iteration with less volatility... Once everything gets locked down, I'll let everyone know. Quote Link to comment Share on other sites More sharing options...
muck Posted April 26, 2011 Author Share Posted April 26, 2011 Quick summary stats on what I think we're going to be using: -2.25 = "strong sell" = 21 weeks = -2.10% / week average ... +799.6% annualized -1.50 = "sell" = 148 weeks = -0.43% / week average ... +34.8% annualized 0.00 = "cash" = 598 weeks = 0.00% / week average ... the market deliverd a -1.5% annualized return during these 598 weeks we're on the sideline 0.75 = "buy" = 404 weeks = +0.43% / week average ... +17.2% annualized 1.50 = "strong buy" = 90 weeks = +0.61% / week average ... +57.4% annualized 2.25 = "really strong buy" = 140 weeks = +1.00% / week average ... +203.7% annualized Annualized return = 28.4% / yr (exclusive of dividends received or paid; interest received on cash balances or paid on leverage; commissions paid; slippage between actual trade levels vs. actual closing prices on Friday; investment management fees; operating fees (audit, etc.)) Quote Link to comment Share on other sites More sharing options...
muck Posted April 29, 2011 Author Share Posted April 29, 2011 It looks like the "original" model is going to be a "+2" for next week ... whereas the "revised" model is in cash (i.e., a reading of "0"). I've not run the analysis that would give any insight as to how to read that ... but, I give it to you nonetheless. REMEMBER --- the "revised" model is in cash about 40% of the time and trades much more frequently than the "original" model. I'll try to run the numbers to see what is to be learned about this sort of combination of signals, but that analysis may not happen today ... caveat emptor ... Quote Link to comment Share on other sites More sharing options...
Ursa Majoris Posted April 29, 2011 Share Posted April 29, 2011 It looks like the "original" model is going to be a "+2" for next week ... whereas the "revised" model is in cash (i.e., a reading of "0"). I've not run the analysis that would give any insight as to how to read that ... but, I give it to you nonetheless. REMEMBER --- the "revised" model is in cash about 40% of the time and trades much more frequently than the "original" model. I'll try to run the numbers to see what is to be learned about this sort of combination of signals, but that analysis may not happen today ... caveat emptor ... There has to be some differentiation in how the input data is being treated or, alternatively is there any data that is only going to one model? Quote Link to comment Share on other sites More sharing options...
muck Posted April 29, 2011 Author Share Posted April 29, 2011 There has to be some differentiation in how the input data is being treated... No, there doesn't. ...alternatively is there any data that is only going to one model? Not in any substantive new data. ************************************************ The original model and the revised models look at the same data, just look at it slightly differently ... which is what's required to get the additional gradations (the original has three outputs and the refined has six). I expect to use one 'trick' the original model uses that the refined model currently does not to see if that makes any difference ... I hope to have more information and/or details up sometime between now and next Friday. Quote Link to comment Share on other sites More sharing options...
Ursa Majoris Posted April 30, 2011 Share Posted April 30, 2011 There has to be some differentiation in how the input data is being treated No, there doesn't. The original model and the revised models look at the same data, just look at it slightly differently ... which is what's required to get the additional gradations (the original has three outputs and the refined has six). I expect to use one 'trick' the original model uses that the refined model currently does not to see if that makes any difference ... Quote Link to comment Share on other sites More sharing options...
muck Posted April 30, 2011 Author Share Posted April 30, 2011 Preliminary modelling indicates that the 'trick' may work for the revised model, too...BUT...it may still not give a different answer as to this coming weeks' scores. Quote Link to comment Share on other sites More sharing options...
muck Posted May 6, 2011 Author Share Posted May 6, 2011 ...in any event, it looks like we'll be long for the coming week...specific reading (and details on the refined model) to come later... Quote Link to comment Share on other sites More sharing options...
muck Posted May 13, 2011 Author Share Posted May 13, 2011 It looks like the refined models are saying "cash" for the coming week. Per my records, there were 65 weeks in the last (nearly) 27yrs with this combination of signals, and while the market was up 60% of the time, the average return for these 65 weeks was exactly 0.0% / week ... meaning, that while you would make money more often than not, when you'd lose money, it'd wipe out the other gains. Quote Link to comment Share on other sites More sharing options...
muck Posted May 13, 2011 Author Share Posted May 13, 2011 (edited) FWIW ... the refined model: Scores are: -1 = short ... 164 weeks (market was down 62% of the time) ... average market return -0.89% / week ... annualized return of the market of 55.4% / year 0 = cash ... 722 weeks (market was up 54% of the time) ... average market return of +0.07% / week ... annualized return of the market of 2.6% / year +1 = long ... 341 weeks (market was up 65% of the time) ... average market return of +0.51% / week ... annualized return of the market of 28.3% / year +2 = more long ... 56 weeks (market was up 66% of the time) ... average market return of +0.83% / week ... annualized return of the market of 52.3% / year +3 = really long ... 113 weeks (market was up 75% of the time) ... average market return of +1.17% / week ... annualized return of the market of 81.3% / year ********************************** Annualized return of +33% / year for nearly 27 years; maximum drawdown was -27% (or about half of the worst drawdown for the S&P500 during the same period). Edited May 20, 2011 by muck Quote Link to comment Share on other sites More sharing options...
muck Posted May 20, 2011 Author Share Posted May 20, 2011 Today's signal will depend on where the market closes. If the market closes below 1338.99, the signal for next week is a "+3" ... and if it's above that, we're in cash. Quote Link to comment Share on other sites More sharing options...
Ursa Majoris Posted May 20, 2011 Share Posted May 20, 2011 Today's signal will depend on where the market closes. If the market closes below 1338.99, the signal for next week is a "+3" ... and if it's above that, we're in cash. So the scale used isn't linear? Because clearly long (+1) and more long (+2) come between really long (+3) and cash (0). Quote Link to comment Share on other sites More sharing options...
muck Posted May 20, 2011 Author Share Posted May 20, 2011 (edited) So the scale used isn't linear? Because clearly long (+1) and more long (+2) come between really long (+3) and cash (0). It's linear in so much that the average P&L from a "-1" week is worse than the average "0" week, which is worse than the average "+1" week, which is worse than the average "+2" week, which is worse than the average "+3" week. The model is constructed off of nearly 100 different inputs and their derivatives. Some times, if one input moves above / below certain levels, there is a compounding effect accross multiple derivatives of that same input, and as a result, give a stronger signal (or reduce a signal), as appropriate. This sort of thing is relatively common in quantitative trading strategies. I think of it as a confidence level in flipping a coin. In this particular case, if we get a "0" score, about 54% of the time the market rises, and the average return is +0.07% for the week ... HOWEVER ... if we get a "+3" score, about 75% of the time the market is up, and the average return for those weeks is +1.17% ... and so, because the odds of being right (i.e., the market going up) are greater than 'normal' and when you're right, you're really right (+1.17% vs. +0.07%), you "bet" a little more on this particular coin flip. Another way to think of it is that, say, you like redheads. You're told you have a blind date with a redhead. You don't know until you walk in the restaurant whether your redhead is Peppermint Patti or Ann Margaret ... or whether the 'redhead' is really more of a strawberry blonde, and you're dinner date is Nicole Kidman. We're about to walk in the restaurant. You'll know what sort of redhead you're looking at based on the closing price of the S&P500. ...I hope that analogy wasn't too obtuse... Edited May 20, 2011 by muck Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.