Holy smokes that's some impressive results! Will be interesting to see how well this ranks, DrKoch might have some competition!
It's also very gratifying to see something good coming out of the Evolver!
It's also very gratifying to see something good coming out of the Evolver!
Rename
Holy smokes is right! Thanks for sharing!
Congratulation to Darcy for taking the top Rankings slot this week!
The way things are going we will need to soon split the Rankings into two - Limit and Non-Limit trading systems, just like in the old days of WL!
The way things are going we will need to soon split the Rankings into two - Limit and Non-Limit trading systems, just like in the old days of WL!
Congratulations!
The NSF ratio for "V Cruiser" is above 4 - so there are 4 times the number of positions rejected compare to those used in the backtest.
When there are many limit candidates on the same date, the Monte Carlo-Lab Same-Date Scramble can give you a distribution of potential outcomes - nothing bad to see here. This was based on a 20-year backtest for the Nasdaq 100.
When there are many limit candidates on the same date, the Monte Carlo-Lab Same-Date Scramble can give you a distribution of potential outcomes - nothing bad to see here. This was based on a 20-year backtest for the Nasdaq 100.
With 11 parameters (the $VIX counts as one) this strategy has a high risk of being over-optimized (even if no classic optimizer was used for its development).
The evolver counts as a "structural" optimizer in my book.
It will be very interesting to see how this strategy fares with new, unseen data i.e. future rankings.
The evolver counts as a "structural" optimizer in my book.
It will be very interesting to see how this strategy fares with new, unseen data i.e. future rankings.
QUOTE:
this strategy has a high risk of being over-optimized
And those auto-generated parameter values like 1.27 or 19.32 point at that, too.
Unfortunately, this strategy has a major problem and reveals a weak spot in block strategies that should be fixed. The same problem can also occur in a C# code strategy, but there it can be more easily avoided.
The second exit rule is to "Sell at Limit 3.94% below Low" when the StochD is below 66.71. Unless the stock gaps down more than 3.94%, this rule will essentially exit the position at the open. That in itself isn't a problem.
The problem is that a limit order with a higher price may be executed first by the strategy. So if price reaches that limit, the position will exit at higher price. Otherwise it will exit at the open. Unfortunately, that's peeking.
Probably WealthLab needs to prune orders in backtests like it does for live trading. I think that would fix it.
The second exit rule is to "Sell at Limit 3.94% below Low" when the StochD is below 66.71. Unless the stock gaps down more than 3.94%, this rule will essentially exit the position at the open. That in itself isn't a problem.
The problem is that a limit order with a higher price may be executed first by the strategy. So if price reaches that limit, the position will exit at higher price. Otherwise it will exit at the open. Unfortunately, that's peeking.
Probably WealthLab needs to prune orders in backtests like it does for live trading. I think that would fix it.
Good catch, Cone!
Bad news for this Strategy, but good news overall for WL8. The Evolver has indeed exposed a subtle flaw in our Backtester! We should be pruning the exit orders, such that if a combination of Market/Limit/Stops are all entered at once, the Backtester keeps only the Market exit. And if multiple Limits/Stops are generated, it selects only one Limit/Stop, the one closest to getting filled. I've made the changes in the Backtester and will re-submit the Rankings after we test the change.
You'd never create a Strategy like this manually, but the Evolver is so randomized that it came up with weird combinations that exposed this flaw. I'll post an update when things are adjusted on site. And look forward to this fix in Build 26 where the Evolver will no longer be able to exploit this.
You'd never create a Strategy like this manually, but the Evolver is so randomized that it came up with weird combinations that exposed this flaw. I'll post an update when things are adjusted on site. And look forward to this fix in Build 26 where the Evolver will no longer be able to exploit this.
QUOTE:
that's peeking.
This evokes some faint memory.
In one of the older WL versions we had something called a "Peek Detector".
This was an overload of some WL library functions.
A BuyAtLimit would mark the complete bar as "touched".
A later BuyAtMarket would trigger some error message because it uses data older than already "touched" i.e. it is peeking.
Same for MarketOnClose orders or any indicator access.
I think it would be relatively simple to implement such a "Peek Detector" functionality in the current C# libs. This could detect even convoluted cases.
Entries (Buy and Short) aren't at question here. If a strategy executes multiple entry orders, they'll all fill up to the buying power.
Glitch's pruning removes the possibility of this type of peeking for exits (already tested for Build 26), so there's no need to try to detect it.
Glitch's pruning removes the possibility of this type of peeking for exits (already tested for Build 26), so there's no need to try to detect it.
A bit disheartening.... but glad the strategy brought some light to flaws early in development. Good work @Cone.
Back to the drawing board for me then...
Let us know what build the evolver is stable, please. Not going to waste too much energy till then.
Back to the drawing board for me then...
Let us know what build the evolver is stable, please. Not going to waste too much energy till then.
It's fixed with Build 26 - possibly later today.
QUOTE:
A bit disheartening
Don't get discouraged!
Development of trading strategies is a journey, there are highs and lows, ups and downs, peaks and valleys...
... I am on this road for 20 years now ... still alive ...
Your Response
Post
Edit Post
Login is required