Well, that's not really true. you have to look more broadly. After Apple's events, stock prices often dip, after they had risen beforehand. Expectations build up, and then are let down.
EXACTLY! They peak (or crest) just before the product announcement, and fall afterwards, I thought this was obvious (although I wouldn't say this happens 100% of the time, perhaps most of the time, or perhaps the probability of exceedance is greater than 50%).
So here's a first thought experiment, take the daily price fluctiation and divide by the daily average trading price (this is a precentage, i. e. yesterday's was about 4%), do this for 5 years of record, form a PDF (obviously it's one-sided), and where would yesterday's 4% fluctuation fall in this distribution, my guess would be a probability of exceedance in excess of 10-20% (maybe higher), so it may not be typical, but it also isn't highly unusual either. That was my point. Now you do lose sight of some key information (i. e. a keynote at 12 noon ET), that would help to explain that day's price fluctuation, however if I were to select 30 shapes (demeaned and normalized (divided by max - min)) that looked somewhat similar to yesterday's (including yesterday's, the others would be unrelated to any preceived Apple influence (granted perhaps this is somewhat subjective (i. e. define Apple influence))) and asked you to pick out yesterday's from that sample, do you really think you could "guess" the correct one greater than random statistical probability? Perhaps you are also good at reading palms?
You mentioned in a follow-on post that companies go from peak-to-peak (crest-to-crest) or trough-to-trough, what do you suppose comes between those crests (troughs!) or troughs (crests!)? Please explain?
I have a great deal of experience WRT ocean (and laboratory) wave time series, have done all kinds of time-domain (up/down/zero crossing) and frequency domain (FFT) analyses, low/high/bandpass filtering of said time series. Harmonic analyses, wave-height distributions (PDF's), wave-height-period (joint PDF) distributions.
So here's an second thought experiment, take 30 NYSE companies each with 5 years of price fluctuation records (say 15 minute samples (or however often the prices are posted)), pass these through a low-pass cutoff filter set to a 1-month cutoff frequency (i. e. eliminate all information greater than 1-month), FFT these filtered time series (or just FFT the raw time series and zero out the frequencies lower than 1/1-month) and look at the resulting spectra. My guess is that the resulting time series might not look significantly different from white noise spectra (broad distribution of price density with no distinct shape, particularly when the shapes are normalized and ensamble averaged). As a follow-on exercise take the price indicies, multiply by the volume traded, again form as a time series, and repeat the above exercise, I have an even greater expectation that the resulting spectra will indeed look like white noise.
So while the term "crapshoot" may not be entirely correct, given that the forcing function is human nature, expectations, and rumors (or lack thereof in Apple's case), I would suggest that this term isn't too far of the mark!
The market ALWAYS punishes Apple after keynotes. It's the basic Apple didn't announce 'widget XXX' knee-jerk reaction. Anyone who knows Apple stock can predict that move with their eyes closed, minus 2-5% can be taken to the bank every time.
The market ALWAYS punishes Apple after keynotes. It's the basic Apple didn't announce 'widget XXX' knee-jerk reaction. Anyone who knows Apple stock can predict that move with their eyes closed, minus 2-5% can be taken to the bank every time.
Agree! But if there's a price runup prior to the keynote, with the expectation (as you've stated a "no brainer") of a reduction afterward, and you bought prior to the runup (on purpose, since you know the pattern a priori), and you sell during the keynote (expecting it to peak), then you take home +2-5%, and that IMHO is what those traderz are doing (buy low sell high). This is a positive feedback mechanism, since once selling begins, the price reduction continues, more people "get out" and the downward trend continues?
EXACTLY! They peak (or crest) just before the product announcement, and fall afterwards, I thought this was obvious (although I wouldn't say this happens 100% of the time, perhaps most of the time, or perhaps the probability of exceedance is greater than 50%).
So here's a first thought experiment, take the daily price fluctiation and divide by the daily average trading price (this is a precentage, i. e. yesterday's was about 4%), do this for 5 years of record, form a PDF (obviously it's one-sided), and where would yesterday's 4% fluctuation fall in this distribution, my guess would be a probability of exceedance in excess of 10-20% (maybe higher), so it may not be typical, but it also isn't highly unusual either. That was my point. Now you do lose sight of some key information (i. e. a keynote at 12 noon ET), that would help to explain that day's price fluctuation, however if I were to select 30 shapes (demeaned and normalized (divided by max - min)) that looked somewhat similar to yesterday's (including yesterday's, the others would be unrelated to any preceived Apple influence (granted perhaps this is somewhat subjective (i. e. define Apple influence))) and asked you to pick out yesterday's from that sample, do you really think you could "guess" the correct one greater than random statistical probability? Perhaps you are also good at reading palms?
You mentioned in a follow-on post that companies go from peak-to-peak (crest-to-crest) or trough-to-trough, what do you suppose comes between those crests (troughs!) or troughs (crests!)? Please explain?
I have a great deal of experience WRT ocean (and laboratory) wave time series, have done all kinds of time-domain (up/down/zero crossing) and frequency domain (FFT) analyses, low/high/bandpass filtering of said time series. Harmonic analyses, wave-height distributions (PDF's), wave-height-period (joint PDF) distributions.
So here's an second thought experiment, take 30 NYSE companies each with 5 years of price fluctuation records (say 15 minute samples (or however often the prices are posted)), pass these through a low-pass cutoff filter set to a 1-month cutoff frequency (i. e. eliminate all information greater than 1-month), FFT these filtered time series (or just FFT the raw time series and zero out the frequencies lower than 1/1-month) and look at the resulting spectra. My guess is that the resulting time series might not look significantly different from white noise spectra (broad distribution of price density with no distinct shape, particularly when the shapes are normalized and ensamble averaged). As a follow-on exercise take the price indicies, multiply by the volume traded, again form as a time series, and repeat the above exercise, I have an even greater expectation that the resulting spectra will indeed look like white noise.
So while the term "crapshoot" may not be entirely correct, given that the forcing function is human nature, expectations, and rumors (or lack thereof in Apple's case), I would suggest that this term isn't too far of the mark!
That's all very interesting, and I'm not about to denigrate you concept. I've done most of the same analysis in my own work when I designed speaker drivers/systems, and other analog and digital devices. So I appreciate the logic.
The problem here is that this is not subject to the same critical analysis. As you obviously know, as long as we have enough data points we can construct a sufficient method to work with it. Not so with human decision making.
I don't believe that we can arrive at a reliable enough method to analyze this, without spending more time that it is worth, unless, we want to come up with something that we can use for our own financial benefit.
One reason why I didn't follow through with my degree in psychology was because I didn't (and still don't) believe that we can understand all of the variables that go into even one persons decisions, much less that of a vast number.
The problem here is that we don't know exactly who is buying and selling at any given time. What percentage of day traders? What percentage of investors such as myself, what percentage of investment houses? these ratios change all of the time, and so will the effects on the market.
That's all very interesting, and I'm not about to denigrate you concept. I've done most of the same analysis in my own work when I designed speaker drivers/systems, and other analog and digital devices. So I appreciate the logic.
The problem here is that this is not subject to the same critical analysis. As you obviously know, as long as we have enough data points we can construct a sufficient method to work with it. Not so with human decision making.
I don't believe that we can arrive at a reliable enough method to analyze this, without spending more time that it is worth, unless, we want to come up with something that we can use for our own financial benefit.
One reason why I didn't follow through with my degree in psychology was because I didn't (and still don't) believe that we can understand all of the variables that go into even one persons decisions, much less that of a vast number.
The problem here is that we don't know exactly who is buying and selling at any given time. What percentage of day traders? What percentage of investors such as myself, what percentage of investment houses? these ratios change all of the time, and so will the effects on the market.
IMHO, I'd agree with most of what you're saying. However, I don't think that has stopped Wall Street from hiring hunderds of PhD's with strong backgrounds in the areas of applied math/physics/engineering (one of my friends here at work was on a PhD student's committee, she was big into numerical modeling of ocean waves, anyway she now works on Wall Street making several times what she could make as either an academic or in the engineering profession). I'd also agree that both thought experiments I suggested above would not bare fruit (i. e. no discernible trends or patterns would emerge (i. e. it would indeed be white noise)), however there are probably other more advanced statistical/mathematical methods that might generate a predictable signal-to-noise ratio (Kalman filtering comes to mind, it's a real time predictor-corrector used in Inertial Navigation Systems (INS), used in all forms of vehicles, and has been shown mathematically to minimize the residual errors). Anytime, you can exceed 50% probability of exceedance (long term), you stand to gain (long term). Although most of human nature may appear to be a random process, I don't think it's a TOTALLY random process, and right off the top of my head, I can't think of an institution that generates more data than the NYSE!
IMHO, I'd agree with most of what you're saying. However, I don't think that has stopped Wall Street from hiring hunderds of PhD's with strong backgrounds in the areas of applied math/physics/engineering (one of my friends here at work was on a PhD student's committee, she was big into numerical modeling of ocean waves, anyway she now works on Wall Street making several times what she could make as either an academic or in the engineering profession). I'd also agree that both thought experiments I suggested above would not bare fruit (i. e. no discernible trends or patterns would emerge (i. e. it would indeed be white noise)), however there are probably other more advanced statistical/mathematical methods that might generate a predictable signal-to-noise ratio (Kalman filtering comes to mind, it's a real time predictor-corrector used in Inertial Navigation Systems (INS), used in all forms of vehicles, and has been shown mathematically to minimize the residual errors). Anytime, you can exceed 50% probability of exceedance (long term), you stand to gain (long term). Although most of human nature may appear to be a random process, I don't think it's a TOTALLY random process, and right off the top of my head, I can't think of an institution that generates more data than the NYSE!
That's true, and it still doesn't work in the long term. Remember a short few years ago there was a fund, I forget the a actual name right now, Long Tern Capital, or something similar. It was doing very well, and had an investment base of a trillion or so dollars that it controlled? It was run by economics PhD's, some of whom had won the Nobel Prize in economics. Do you remember the sudden collapse? Just one bad investment, and the entire thing came crashing down. It was thought that it would severely damage the entire financial community. There was panic before it was restructured.
The only way this works is to do the analysis manually. We can pick patterns out of the noise better than most any algorythms at this time when the patterns are made by human decision making. All of the physical problems are understandable, if they are not too complex. Some of them might never be subject to analysis, such as chaotic systems, where one small push can upset the entire balance, and send it into another direction. human decisions are often like that. There isn't rational thought behind all of these decisions, though at the institutional level, things are usually calmer.
But, then, some of it is pretty simple as well. I know some of those patterns. If Apple comes up with what the market expects, the stock will continue to go up afterwards. I'm not talking here about what the price will be a couple of weeks later, because the price at that time will be distorted by other market forces not directly related to Apple, except in the most general way.
Comments
Well, that's not really true. you have to look more broadly. After Apple's events, stock prices often dip, after they had risen beforehand. Expectations build up, and then are let down.
EXACTLY! They peak (or crest) just before the product announcement, and fall afterwards, I thought this was obvious (although I wouldn't say this happens 100% of the time, perhaps most of the time, or perhaps the probability of exceedance is greater than 50%).
So here's a first thought experiment, take the daily price fluctiation and divide by the daily average trading price (this is a precentage, i. e. yesterday's was about 4%), do this for 5 years of record, form a PDF (obviously it's one-sided), and where would yesterday's 4% fluctuation fall in this distribution, my guess would be a probability of exceedance in excess of 10-20% (maybe higher), so it may not be typical, but it also isn't highly unusual either. That was my point. Now you do lose sight of some key information (i. e. a keynote at 12 noon ET), that would help to explain that day's price fluctuation, however if I were to select 30 shapes (demeaned and normalized (divided by max - min)) that looked somewhat similar to yesterday's (including yesterday's, the others would be unrelated to any preceived Apple influence (granted perhaps this is somewhat subjective (i. e. define Apple influence))) and asked you to pick out yesterday's from that sample, do you really think you could "guess" the correct one greater than random statistical probability? Perhaps you are also good at reading palms?
You mentioned in a follow-on post that companies go from peak-to-peak (crest-to-crest) or trough-to-trough, what do you suppose comes between those crests (troughs!) or troughs (crests!)? Please explain?
I have a great deal of experience WRT ocean (and laboratory) wave time series, have done all kinds of time-domain (up/down/zero crossing) and frequency domain (FFT) analyses, low/high/bandpass filtering of said time series. Harmonic analyses, wave-height distributions (PDF's), wave-height-period (joint PDF) distributions.
So here's an second thought experiment, take 30 NYSE companies each with 5 years of price fluctuation records (say 15 minute samples (or however often the prices are posted)), pass these through a low-pass cutoff filter set to a 1-month cutoff frequency (i. e. eliminate all information greater than 1-month), FFT these filtered time series (or just FFT the raw time series and zero out the frequencies lower than 1/1-month) and look at the resulting spectra. My guess is that the resulting time series might not look significantly different from white noise spectra (broad distribution of price density with no distinct shape, particularly when the shapes are normalized and ensamble averaged). As a follow-on exercise take the price indicies, multiply by the volume traded, again form as a time series, and repeat the above exercise, I have an even greater expectation that the resulting spectra will indeed look like white noise.
So while the term "crapshoot" may not be entirely correct, given that the forcing function is human nature, expectations, and rumors (or lack thereof in Apple's case), I would suggest that this term isn't too far of the mark!
The market ALWAYS punishes Apple after keynotes. It's the basic Apple didn't announce 'widget XXX' knee-jerk reaction. Anyone who knows Apple stock can predict that move with their eyes closed, minus 2-5% can be taken to the bank every time.
Agree! But if there's a price runup prior to the keynote, with the expectation (as you've stated a "no brainer") of a reduction afterward, and you bought prior to the runup (on purpose, since you know the pattern a priori), and you sell during the keynote (expecting it to peak), then you take home +2-5%, and that IMHO is what those traderz are doing (buy low sell high). This is a positive feedback mechanism, since once selling begins, the price reduction continues, more people "get out" and the downward trend continues?
Is that what you meant?
The shipping delay is probably to get Tiger Server for Intel out for general seeding/testing.
You can buy the Universal version of Mac OS X Server today, so that's not the reason for the dealy.
PS: Sorry for actually discussing on topic stuff
EXACTLY! They peak (or crest) just before the product announcement, and fall afterwards, I thought this was obvious (although I wouldn't say this happens 100% of the time, perhaps most of the time, or perhaps the probability of exceedance is greater than 50%).
So here's a first thought experiment, take the daily price fluctiation and divide by the daily average trading price (this is a precentage, i. e. yesterday's was about 4%), do this for 5 years of record, form a PDF (obviously it's one-sided), and where would yesterday's 4% fluctuation fall in this distribution, my guess would be a probability of exceedance in excess of 10-20% (maybe higher), so it may not be typical, but it also isn't highly unusual either. That was my point. Now you do lose sight of some key information (i. e. a keynote at 12 noon ET), that would help to explain that day's price fluctuation, however if I were to select 30 shapes (demeaned and normalized (divided by max - min)) that looked somewhat similar to yesterday's (including yesterday's, the others would be unrelated to any preceived Apple influence (granted perhaps this is somewhat subjective (i. e. define Apple influence))) and asked you to pick out yesterday's from that sample, do you really think you could "guess" the correct one greater than random statistical probability? Perhaps you are also good at reading palms?
You mentioned in a follow-on post that companies go from peak-to-peak (crest-to-crest) or trough-to-trough, what do you suppose comes between those crests (troughs!) or troughs (crests!)? Please explain?
I have a great deal of experience WRT ocean (and laboratory) wave time series, have done all kinds of time-domain (up/down/zero crossing) and frequency domain (FFT) analyses, low/high/bandpass filtering of said time series. Harmonic analyses, wave-height distributions (PDF's), wave-height-period (joint PDF) distributions.
So here's an second thought experiment, take 30 NYSE companies each with 5 years of price fluctuation records (say 15 minute samples (or however often the prices are posted)), pass these through a low-pass cutoff filter set to a 1-month cutoff frequency (i. e. eliminate all information greater than 1-month), FFT these filtered time series (or just FFT the raw time series and zero out the frequencies lower than 1/1-month) and look at the resulting spectra. My guess is that the resulting time series might not look significantly different from white noise spectra (broad distribution of price density with no distinct shape, particularly when the shapes are normalized and ensamble averaged). As a follow-on exercise take the price indicies, multiply by the volume traded, again form as a time series, and repeat the above exercise, I have an even greater expectation that the resulting spectra will indeed look like white noise.
So while the term "crapshoot" may not be entirely correct, given that the forcing function is human nature, expectations, and rumors (or lack thereof in Apple's case), I would suggest that this term isn't too far of the mark!
That's all very interesting, and I'm not about to denigrate you concept. I've done most of the same analysis in my own work when I designed speaker drivers/systems, and other analog and digital devices. So I appreciate the logic.
The problem here is that this is not subject to the same critical analysis. As you obviously know, as long as we have enough data points we can construct a sufficient method to work with it. Not so with human decision making.
I don't believe that we can arrive at a reliable enough method to analyze this, without spending more time that it is worth, unless, we want to come up with something that we can use for our own financial benefit.
One reason why I didn't follow through with my degree in psychology was because I didn't (and still don't) believe that we can understand all of the variables that go into even one persons decisions, much less that of a vast number.
The problem here is that we don't know exactly who is buying and selling at any given time. What percentage of day traders? What percentage of investors such as myself, what percentage of investment houses? these ratios change all of the time, and so will the effects on the market.
That's all very interesting, and I'm not about to denigrate you concept. I've done most of the same analysis in my own work when I designed speaker drivers/systems, and other analog and digital devices. So I appreciate the logic.
The problem here is that this is not subject to the same critical analysis. As you obviously know, as long as we have enough data points we can construct a sufficient method to work with it. Not so with human decision making.
I don't believe that we can arrive at a reliable enough method to analyze this, without spending more time that it is worth, unless, we want to come up with something that we can use for our own financial benefit.
One reason why I didn't follow through with my degree in psychology was because I didn't (and still don't) believe that we can understand all of the variables that go into even one persons decisions, much less that of a vast number.
The problem here is that we don't know exactly who is buying and selling at any given time. What percentage of day traders? What percentage of investors such as myself, what percentage of investment houses? these ratios change all of the time, and so will the effects on the market.
IMHO, I'd agree with most of what you're saying. However, I don't think that has stopped Wall Street from hiring hunderds of PhD's with strong backgrounds in the areas of applied math/physics/engineering (one of my friends here at work was on a PhD student's committee, she was big into numerical modeling of ocean waves, anyway she now works on Wall Street making several times what she could make as either an academic or in the engineering profession). I'd also agree that both thought experiments I suggested above would not bare fruit (i. e. no discernible trends or patterns would emerge (i. e. it would indeed be white noise)), however there are probably other more advanced statistical/mathematical methods that might generate a predictable signal-to-noise ratio (Kalman filtering comes to mind, it's a real time predictor-corrector used in Inertial Navigation Systems (INS), used in all forms of vehicles, and has been shown mathematically to minimize the residual errors). Anytime, you can exceed 50% probability of exceedance (long term), you stand to gain (long term). Although most of human nature may appear to be a random process, I don't think it's a TOTALLY random process, and right off the top of my head, I can't think of an institution that generates more data than the NYSE!
IMHO, I'd agree with most of what you're saying. However, I don't think that has stopped Wall Street from hiring hunderds of PhD's with strong backgrounds in the areas of applied math/physics/engineering (one of my friends here at work was on a PhD student's committee, she was big into numerical modeling of ocean waves, anyway she now works on Wall Street making several times what she could make as either an academic or in the engineering profession). I'd also agree that both thought experiments I suggested above would not bare fruit (i. e. no discernible trends or patterns would emerge (i. e. it would indeed be white noise)), however there are probably other more advanced statistical/mathematical methods that might generate a predictable signal-to-noise ratio (Kalman filtering comes to mind, it's a real time predictor-corrector used in Inertial Navigation Systems (INS), used in all forms of vehicles, and has been shown mathematically to minimize the residual errors). Anytime, you can exceed 50% probability of exceedance (long term), you stand to gain (long term). Although most of human nature may appear to be a random process, I don't think it's a TOTALLY random process, and right off the top of my head, I can't think of an institution that generates more data than the NYSE!
That's true, and it still doesn't work in the long term. Remember a short few years ago there was a fund, I forget the a actual name right now, Long Tern Capital, or something similar. It was doing very well, and had an investment base of a trillion or so dollars that it controlled? It was run by economics PhD's, some of whom had won the Nobel Prize in economics. Do you remember the sudden collapse? Just one bad investment, and the entire thing came crashing down. It was thought that it would severely damage the entire financial community. There was panic before it was restructured.
The only way this works is to do the analysis manually. We can pick patterns out of the noise better than most any algorythms at this time when the patterns are made by human decision making. All of the physical problems are understandable, if they are not too complex. Some of them might never be subject to analysis, such as chaotic systems, where one small push can upset the entire balance, and send it into another direction. human decisions are often like that. There isn't rational thought behind all of these decisions, though at the institutional level, things are usually calmer.
But, then, some of it is pretty simple as well. I know some of those patterns. If Apple comes up with what the market expects, the stock will continue to go up afterwards. I'm not talking here about what the price will be a couple of weeks later, because the price at that time will be distorted by other market forces not directly related to Apple, except in the most general way.
This article I think most will appreciate. It starts with the machines, and then goes to a good breakdown of the software.
His insights are very interesting. The article is 5 pages.
I just posted this to the Leopard forum.
http://www.computerworld.com/action/...icleId=9002267
Yes!
http://hardware.seekingalpha.com/article/15126