good stocks to invest in?

[CPA22]

the market is set for a correction soon. Be very careful where you invest.

 

[collegiateLifter]

CollegiateLifter - I didn't say anything about the impossibility of multi-sigma events - just the low probability.
There is a big difference.
Many of the issues with the probability theory is whether you are treating the standard deviation as one over a normal Guassian curve - which is an inappropriate thing for application against the stock market since it won't deviate into the negative (if you are looking at the change in value, then it is more acceptable, but if you are looking at prices, they can't go negative, they just delist).
When you don't apply it to the proper curve (I'm blanking on the proper curve term name at the moment - I want to say lepto something, but without looking it up I am sure I am wrong on that) - then you have a more accurate spread - but still can never be totally on with the predictions of the next sigma event.

In the case of LTCM there were 'events' that should have never occured in the thousands of years on earth, but did nevertheless....

Anyway, all of these theoretical applications do rely on the efficient market hypothesis, right?

 

[OMGWTFBBQ]

In the case of LTCM there were 'events' that should have never occured in the thousands of years on earth, but did nevertheless....

Anyway, all of these theoretical applications do rely on the efficient market hypothesis, right?

Of the theoretical applications that I have mentioned, none rely on the EMH. They are all based on the math involved with analysis of chaotic events.

As for the LTCM stuff - I know in multiple books they have referred to what happened as a 20 sigma event or even higher, but I have seen pretty good arguments made that the calculations done to get that figure were very wrong.
Off the top of my head I could see one immediate issue being if you calculated your standard deviation over a Guassian curve and you didn't look at the data properly related to that.
Financial data shouldn't be (can't be) mapped to a Guassian curve - the change in movements can be - but the straight up figures can't be.
Using the wrong distribution curve is one way to see multi-sigma events occurring more often than they should in your model.
Again - I am just thinking off the top of my head right now, I don't recall where I read it and I'm not up on the modeling that they did.

Generally speaking, an individual investor that is playing with less than $200K doesn't really have to worry as much about the same issues that funds have to worry about - the impact of your buying doesn't change the movement of the stock the same way that a multi-million dollar fund entering/existing a stock does. (not to mention that hedge funds usually do more complicated things than the usual individual investor)

 

[Mandinka2]

Not to upset you OMG but I read a recent report on problems with applying Brownian motion type models to share prices - actuaries are mostly developing Markov regime switching models right now. Do your models build in allowances for volatility changes over time ? - there is a lot of empirical evidence to suggest that we see different volatility regimes - this is consitent with the rotational investment strategy using different fund managers depending on whether we are in a bear (choose value fund managers) or bull (growth fund managers) market. I'll try and dig it up and post up some of the results but there does appear to exist a high degree of dependence.
I'm finally curious as to the base assumption to your models , are they momentum based (firms which performed well in the past are likely to do so again in the future?).

 

[OMGWTFBBQ]

Not to upset you OMG but I read a recent report on problems with applying Brownian motion type models to share prices - actuaries are mostly developing Markov regime switching models right now. Do your models build in allowances for volatility changes over time ? - there is a lot of empirical evidence to suggest that we see different volatility regimes - this is consitent with the rotational investment strategy using different fund managers depending on whether we are in a bear (choose value fund managers) or bull (growth fund managers) market. I'll try and dig it up and post up some of the results but there does appear to exist a high degree of dependence.
I'm finally curious as to the base assumption to your models , are they momentum based (firms which performed well in the past are likely to do so again in the future?).

Agreed - and I thought I was making that point - the Brownian motion can't be applied directly to share prices due to an incorrect assumption in that model that they would fall on a Guassian curve, which is not appropriate in that case. There is a curve on which they can be modeled more appropriately, but in the end, one is better off not using the prices directly.
You can use the same technique instead on the change in movement of them and then Guassian curves are technically more appropriate.

To be fair, that is a more basic approach and a lot of the stuff going on uses data that I don't current have access to - so this is sort of a dumbed down version of the "real thing" that funds are using. A closer representation to what they are really tracking would be change in market cap.

I should also point out that I don't have a single set of code that I just look at - I have a series of them. There is the Brownian motion stuff (which is newer to me on the side of financial application), there is the Markov style (which has a few different approaches, some of which are more arguably Bayesian I guess), there are the neural nets (which my definition you don't truly know what they are doing inside), and then there are genetic algorithms to choose the most effective technical analysis combination for each give stock over a fixed time series dataset in the past.
Then there are a series of scripts that just perform basic technical analysis - the equivalent of reading charts, but automated and on a large scale so that you get a subset of the full market as a suggestion for further review.

As for what the code does - each of them look for different things, in the end they are looking for what they feel will move up/down in N days ahead, but they do it with different strategies in the end.
But it is nice when they converge and agree that something is about to move up/down - that is reassuring.
The most recent example of that was a few days ago when they all pointed to TRO.

As for the momentum approach - they don't all look at that sort of thing - but if you believe in any of the chaotic modeling and/or Hurst - that would show that there is a definitely bias so that whatever happened "yesterday" (meaning in recent previous time series events, not necessarily a full day, but in recent time units, whether it is seconds, minutes, days, years, etc), has a higher than random chance of happening again "today".
And this is related - as you say - to among other things the change in movement and the emotional aspect as news spreads over time through the investment pool of potential buyers/sellers. A chain reaction of sorts.

I'm not sure if that answered your question or not.

 

[Mandinka2]

I'm not sure if that answered your question or not.

Better than I expressed it , you should get that job.