Math/Stats theories sharing

Quote from j2ee:

welcome to prove it.
More...

He doesn't have to. Humpy's attitude is similar to other professionals' experience that since not everybody can time markets, then it's impossible, but that since not everybody can anybody who does usually can with small potatoes.

If you can with big potatoes, you're probably not the target of Humpy's or anyone else like thems' attitude.

All of those strategies were written years ago, and that site will be taken down eventually, so while you're at it, that is the only place that you'll find information about what the OP is looking for.

Most of the top 10's in those categories are mine, so to say I'm more than familiar with every strategy in those sections is a little understated.

OP:

If you're objective is:

-High Win Percentage

Lower your stops to be almost non-existant, use frequent semi-random or nonsemi-random entries

-High Win/Loss Ratio

Optimizing Gentically with or without Sharpe Ratio sorting will yield the highest w/l

-Highest Net Profit

Choose the strategy with the highest number of trades and largest average win.

(This usually involves a combination of the previous two and only by choosing the optimizations based solely on this criteria will you find that the highest np usually overtrades no matter how high you set your slippage and commission settings).

-Best Average Monte Carlo

The Best Average Monte Carlo simulation assumes you have software that will do this for you, but since MC is to assess hiden risk, the use of MC as an optimization criteria is not something I've heard can bear out the most optimal trading parameters of a system since MC is used to determine whether tail risk exists in the current system. MC, therefore, is not used to optimze results as much as it is to determine the probability of "black swans."

-Highest Sharpe

The program's optimization parameter will function exactly as the highest net profit, only the effect usually is a slightly lower w/l ratio value than the highest w/l, and a return to max dd that's usually the highest of all these optimization criteria.
 
Quote from outsource100:

The real truth remains unspoken
More...

The real truth:

There is never enough data, and your mc sim only tells you the probability in that period.

My counterargument is usually that the data we have came to exist in its current format between 97-99, so it isn't like if you don't have datasets with those black swans in them, I don't have this problem because I test futures mostly and that means that the small micro events that lead to big blow ups are even further and farther between because market crashes happen less often than collapses in individual stocks.

The black swan of the market is irrelevant to futures trading if you got risk management, but large positions in individual stocks is where most of the usefullness of MC lies.

The "usefullness" only plays out if you make a decision not to use the system, and it eventually blows up. If a number makes you sleep good the probability of profit for me when I analyzed some of my strategies was so high that I consider it to mean that my system is stable, and not necessarily that it won't ever lose or lose big.

The probabilities from the MC just help me sleep at night, and to some extent, that is all I can say about the times I've analyzed my Monte Carlo sims. Simulated result analysis in MC should be substantially similar after a certain number of live trades. If they're significantly different, this can indicate a parameter may not have fitted the data as well as the previous simulations suggest, so that parameter or some logical function in the system should be modified.
 
Noone that I have ever heard of has really cracked the market forecasting accurately problem. The solution should work across all markets all the time.

When or if someone ever does he/she will own most of the planet's wealth before ruining the markets. i.e. others catch onto what he/she has accomplished and not bother anymore.
 
Quote from Humpy:

Noone that I have ever heard of has really cracked the market forecasting accurately problem. The solution should work across all markets all the time.

When or if someone ever does he/she will own most of the planet's wealth before ruining the markets. i.e. others catch onto what he/she has accomplished and not bother anymore.
More...

...........



I can't seem to find the cricket chirping sound



Did I mention that the simulations I have go to 10 to the 14th?


By my estimates, this pays the national debt, and I only trade 4 markets.

The "others catch onto"... assumption is false.

Nobody can duplicate those systems, and the reverse engineering theories are only that.

There's a stack of code, and if you'd like to trade the way that code does....um, yeah, there's no way you can unless you whimsically feel that you have something that works as well as this.
 
Quote from bwolinsky:

The real truth:

There is never enough data, and your mc sim only tells you the probability in that period.

My counterargument is usually that the data we have came to exist in its current format between 97-99, so it isn't like if you don't have datasets with those black swans in them, I don't have this problem because I test futures mostly and that means that the small micro events that lead to big blow ups are even further and farther between because market crashes happen less often than collapses in individual stocks.

The black swan of the market is irrelevant to futures trading if you got risk management, but large positions in individual stocks is where most of the usefullness of MC lies.

The "usefullness" only plays out if you make a decision not to use the system, and it eventually blows up. If a number makes you sleep good the probability of profit for me when I analyzed some of my strategies was so high that I consider it to mean that my system is stable, and not necessarily that it won't ever lose or lose big.

The probabilities from the MC just help me sleep at night, and to some extent, that is all I can say about the times I've analyzed my Monte Carlo sims. Simulated result analysis in MC should be substantially similar after a certain number of live trades. If they're significantly different, this can indicate a parameter may not have fitted the data as well as the previous simulations suggest, so that parameter or some logical function in the system should be modified.
More...

Not never enough,but always different,although there are similarities.The attempt to always stay on the correct side,can be very time consuming,labor intensive and more so, pretty expensive.Detecting the 'swans' is a completely separate part of the monitoring and analysis.In 99% when it happens,it usually gives everything back next day.
 
Quote from bwolinsky:

He doesn't have to. Humpy's attitude is similar to other professionals' experience that since not everybody can time markets, then it's impossible, but that since not everybody can anybody who does usually can with small potatoes.

If you can with big potatoes, you're probably not the target of Humpy's or anyone else like thems' attitude.

All of those strategies were written years ago, and that site will be taken down eventually, so while you're at it, that is the only place that you'll find information about what the OP is looking for.

Most of the top 10's in those categories are mine, so to say I'm more than familiar with every strategy in those sections is a little understated.

OP:

If you're objective is:

-High Win Percentage

Lower your stops to be almost non-existant, use frequent semi-random or nonsemi-random entries

-High Win/Loss Ratio

Optimizing Gentically with or without Sharpe Ratio sorting will yield the highest w/l

-Highest Net Profit

Choose the strategy with the highest number of trades and largest average win.

(This usually involves a combination of the previous two and only by choosing the optimizations based solely on this criteria will you find that the highest np usually overtrades no matter how high you set your slippage and commission settings).

-Best Average Monte Carlo

The Best Average Monte Carlo simulation assumes you have software that will do this for you, but since MC is to assess hiden risk, the use of MC as an optimization criteria is not something I've heard can bear out the most optimal trading parameters of a system since MC is used to determine whether tail risk exists in the current system. MC, therefore, is not used to optimze results as much as it is to determine the probability of "black swans."

-Highest Sharpe

The program's optimization parameter will function exactly as the highest net profit, only the effect usually is a slightly lower w/l ratio value than the highest w/l, and a return to max dd that's usually the highest of all these optimization criteria.
More...

everything is easy when it is only in backtest data. Diffculty comes from if the predictive power holds in future or not. Sharpe ratio cannot be even + if the system keeps losing money.
 
Quote from j2ee:

everything is easy when it is only in backtest data. Diffculty comes from if the predictive power holds in future or not. Sharpe ratio cannot be even + if the system keeps losing money.
More...

You wouldn't trade a losing system, doesn't this debate assume that it is profitable in terms of producing at least somewhat above 60% monthly win percentages?
 
Quote from bwolinsky:

You wouldn't trade a losing system, doesn't this debate assume that it is profitable in terms of producing at least somewhat above 60% monthly win percentages?
More...

You make big claims - would you care to substantiate them ?
 
You must log in or register to reply here.
Back
Top