Index | Recent Threads | Who's Online | Search

Posts: 14    Pages: 2    Prev 1 2
Last Post
This topic has been viewed 501 times and has 13 replies
Chipper6
Re: Limits raised for simulation and live strategy positions

I do not see how the recent shifts in the market from small-caps to large-caps (mega-caps?) wouldn’t completely skew the results of any machine-learning/AI/statistical methods.
That's just it! With the equal weight benchmarks composed of the same universe as your model, you can detect a shift in the market and pull the plug on a model before losses pile up. I have been doing a lot of that over the past two years, but I would have had much better results if this platform would have built-in support for it.

Yes, I can pull the data into Excel. But it's too much work and therefore not practical for most of us to do that on a regular basis.

Dec 23, 2020 8:34:21 AM       
yuvaltaylor
Re: Limits raised for simulation and live strategy positions

Thanks for the comments.

In the short term, we will be adding the following benchmarks to our Research list:

S&P Small Cap (IJR)
S&P Mid Cap (IJH)
NASDAQ 100 3X Long (TQQQ)
NASDAQ 100 3X Short (SQQQ)
VIX (VXX)
Russell 2000 Value (IWN)
Russell 2000 Growth (IWO)

Custom benchmarks is a longer-term project. Here are five possibilities for how they could work. What are your thoughts on these?

1. At the same time as the engine runs the screen/ranking system/simulation, it would generate a series based on the universe in use (equal weight) with quarterly rebalancing and no slippage.

2. Allow users to use the equity curve of another screen backtest as a benchmark. Users could then create any benchmark they want by running a screen backtest. This immediately presents problems, though, for live strategies, as screens are not automatically updated.

3. Same as above but using aggregate series. This is much less problematic since series are automatically updated when used in a live strategy.

4. Create a new tool that enables users to generate custom benchmarks. This would be the most work for us and take the longest.

5. Enable all ETF tickers as benchmarks. This would probably be the easiest to do.

I'm favoring #1, but please let me know your thoughts. If #2 or #3 were done, then there would be error messages anytime you tried to run a screen or simulation and your benchmark dates weren't covered (i.e. if your screen started in 2005 and your aggregate series started in 2007).

Personally, for years I've been pulling screen results into Excel to use as benchmarks. It's a bit of a hassle, I admit.

Yuval Taylor
Product Manager, Portfolio123
invest(igations)
Any opinions or recommendations in this message are not opinions or recommendations of Portfolio123 Securities LLC.

Dec 23, 2020 9:25:42 AM       
Jrinne
Re: Limits raised for simulation and live strategy positions


1. At the same time as the engine runs the screen/ranking system/simulation, it would generate a series based on the universe in use (equal weight) with quarterly rebalancing and no slippage.

Yuval,

This sees almost perfect.

Honest question: how much does one lose with the quarterly rebalance? Seems like there might be some stocks that move in or out of the universe that you do not pick up immediately but shouldn’t it track a weekly rebalance closely? For stocks that remain in the unverse a weekly or quarterly rebalance makes no difference?

This would track something like SP500 or Russell 3000 (not based on liquidity or volume) even closer than my usual universe—missing the entry of Tesla by a limited number of weeks perhaps.

I would be interested in what Chaim and others say but I think this might be me preference for machine-learning/AI on first blush.

Thanks.

Jim

From time to time you will encounter Luddites, who are beyond redemption.
--de Prado, Marcos López on the topic of machine learning for financial applications

Dec 23, 2020 9:43:32 AM       
Edit 1 times, last edit by Jrinne at Dec 23, 2020 9:44:33 AM
Chipper6
Re: Limits raised for simulation and live strategy positions

Custom benchmarks is a longer-term project. Here are five possibilities for how they could work. What are your thoughts on these?

1. At the same time as the engine runs the screen/ranking system/simulation, it would generate a series based on the universe in use (equal weight) with quarterly rebalancing and no slippage.

2. Allow users to use the equity curve of another screen backtest as a benchmark. Users could then create any benchmark they want by running a screen backtest. This immediately presents problems, though, for live strategies, as screens are not automatically updated.

3. Same as above but using aggregate series. This is much less problematic since series are automatically updated when used in a live strategy.

4. Create a new tool that enables users to generate custom benchmarks. This would be the most work for us and take the longest.

5. Enable all ETF tickers as benchmarks. This would probably be the easiest to do.

I'm favoring #1, but please let me know your thoughts. If #2 or #3 were done, then there would be error messages anytime you tried to run a screen or simulation and your benchmark dates weren't covered (i.e. if your screen started in 2005 and your aggregate series started in 2007).

Personally, for years I've been pulling screen results into Excel to use as benchmarks. It's a bit of a hassle, I admit.
Yuval, My preference is #1 as well. But if #2, #3, or #4 are easier and will get done faster than by all means do those.

The main thing is to get it done.

Dec 23, 2020 10:41:42 AM       
Posts: 14    Pages: 2    Prev 1 2
 Last Post