time series rank of prices, pls help refine my trading rule

[rs2000]

Hi all,

I am backtesting my Alpha below where weights in the stocks are defined as below,
I rank the prices of last 1 year from 0 to 1.

if rank(closeprice,250) >0.1 then 1- rank(closeprice,250)



The simulator will use the above expression everyday to weigh stocks and eventually will be subindustry neutralized.

My rationale is that as the price moves to the high prices in last 1 year, its more likely to drop and revert to mean so 1- rank(closeprices,250) will give is smaller weight.
Can you please give your thoughts on the above, its not giving good results and the sharp ratio is close to 0.2 only
I want to use the time series rank of prices and generate something with it.

Please help. thanks!

 

[conduit]

What makes you think this assertion alone is valid? When you look at stocks you hardly ever see a stock price mean reverting near last years highest prices. It either maintains momentum and keep on rising or it corrected earlier. My point being is that stock prices today and the investors behind driving stop prices could not care less about last year's high or lows.

Hi all,
My rationale is that as the price moves to the high prices in last 1 year, its more likely to drop and revert to mean so ...

Please help. thanks!

 

[rs2000]

thanks I do realize my thoughts had shortcomings
I was trying to use the time series of prices and trying to implement mean reversion strategies there. Looks like I could shorten the time series to (lets says) a month? ranking the current price in the last one month time series could help identify the probability of a mean reversal? I am looking for start here and will add on moving averages and other stuff
thanks again

 

[conduit]

That makes a lot more sense but all subject to whether you can validate such assumptions. If you can identify stocks that subject their return profile to historic price levels then you are already a step further. I myself am not a believer in such assumptions but do whatever works for you

thanks I do realize my thoughts had shortcomings
I was trying to use the time series of prices and trying to implement mean reversion strategies there. Looks like I could shorten the time series to (lets says) a month? ranking the current price in the last one month time series could help identify the probability of a mean reversal? I am looking for start here and will add on moving averages and other stuff
thanks again

 

[Simples]

To explore mean reversion you could check out bollinger bands. They make it easy to generate "signals", ie. track movement from outside band to the inside. That alone is hardly enough, but a good place to start as any.

 

[conduit]

so, when will you put hard cold cash on the line, when prices touch 1 standard deviation bands? 2? 3? I properly tested most technical indicators a long time ago and even then almost all completely fell apart and did not hold up to unbiased, rigorous testing. If you believe they work how do you use them?

To explore mean reversion you could check out bollinger bands. They make it easy to generate "signals", ie. track movement from outside band to the inside. That alone is hardly enough, but a good place to start as any.

 

[Simples]

so, when will you put hard cold cash on the line, when prices touch 1 standard deviation bands? 2? 3? I properly tested most technical indicators a long time ago and even then almost all completely fell apart and did not hold up to unbiased, rigorous testing. If you believe they work how do you use them?

If price touches the bands, I'd naively, if disregarding main trend, anticipate some sort of continuation of PA, not reversal. Though I can understand that newbies would hope to expect reversal at ie. 3-4 SD, and go long directly there with infinite position sizing.

What do you mean "properly tested most technical indicators"? Nobody are saying any TA indicator works by themselves, and if you're attempting to naively brute force backtesting of each "TA indicator", it's not surprising you don't find success there. Please read the articles on stockcharts.com, ie. bollinger bands:

As with other indicators,Bollinger Bands are not meant to be used as a stand alone tool. Chartists should combine Bollinger Bands with basic trend analysis and other indicators for confirmation.​

This goes for most "TA indicators" and formulas. By themselves, they do not provide trading setups or risk management at all.

I don't see why any TA indicator provide any profitable signals by themselves, but I do believe people can use them skillfully to become profitable.

 

[conduit]

what I mean is that I also ran through various combinations but to be honest it was grunt work and I really hated to do it because: I simply do not believe that algorithms that are by definition lagging shall have any predictive power. It is a balancing act to even hope that any combination of lagging indicators, weighted in any desirable fashion, would improve predictive power. the comment from "stockcharts" you posted makes very little sense, think about it: When you use bollinger bands one takes a mean-reversion approach, obviously. Now how does a "basic trend analysis" fit into this question. There is obviously no established trend when we are looking for a mean-reversion setup. Its perfectly apparent that such is written by someone who has never worked as a career choice as trader or investor. I cannot and have no inclination to disprove technical analysis but when I hear bogus claims or confused people talking about TA then I sometimes cannot help it to point out facts that simply contradict what has been said. In this particular example I am not talking about you but the referenced text.

If price touches the bands, I'd naively, if disregarding main trend, anticipate some sort of continuation of PA, not reversal. Though I can understand that newbies would hope to expect reversal at ie. 3-4 SD, and go long directly there with infinite position sizing.

What do you mean "properly tested most technical indicators"? Nobody are saying any TA indicator works by themselves, and if you're attempting to naively brute force backtesting of each "TA indicator", it's not surprising you don't find success there. Please read the articles on stockcharts.com, ie. bollinger bands:

As with other indicators,Bollinger Bands are not meant to be used as a stand alone tool. Chartists should combine Bollinger Bands with basic trend analysis and other indicators for confirmation.​

This goes for most "TA indicators" and formulas. By themselves, they do not provide trading setups or risk management at all.

I don't see why any TA indicator provide any profitable signals by themselves, but I do believe people can use them skillfully to become profitable.

 

[Simples]

I simply do not believe that algorithms that are by definition lagging shall have any predictive power.

Totally agree. Only price is the true master and cannot be predicted 100%, except in hindsight by "curve fitting". However, prediction and finding possible high-probability trade setups are two different beasts altogether. You only need a couple of setups during a typical chart. Shouldn't need to predict every twist and turn of price. In a longer trend, you only need good entries that are easy to defend and have meaningful impact on PnL. Sounds easy when I say it, but very hard to do consistently, precisely because of your point above.