Introducing a new system "CrystalBall" and would appreciate feedback

Trading and Markets
1

Good day everyone.

I am new to C2 and as I figure out a few details on C2, would really appreciate any feedback on my trading strategy Q4.CrystalBall.

Link to Q4.CrystalBall on C2:

Strategy and back test details from 2004 - Jun 2018:

Currently I have not yet set up the system to charge while it is being tried out by C2 community.

This is not a get rich soon system but rather focuses on long term wealth building with Market beating returns and lower draw downs.

System can be traded in retirement accounts given that this is a long only system on Stocks.

Some of the key strategy details are:

  1. Based on testing from from 2004 - Jun 2018
    35% annual return
    15% max monthly drawdown
    1.57 sharp and a 0.55 correlation to S&P
    This period covers all different market environments and the system behavior can be seen in the charts provided at the link above.

  2. Strategy trades High Growth large cap companies and holds 5 stocks at any given time.

  3. Trades are taken at the end of month and held for the next month.

  4. Position sizing is based on current market environment. Trades are generated based on price, volatility and correlations. System sidesteps down markets and moves to cash to preserve capital.

  5. Only a portion of the Capital is invested at any given time.

  6. Any large losses are prevented by using large cap stocks, with position sizing and rebalancing monthly to identify the stocks that are right for the current market environment.

  7. For the 14 year period from 2004-2018:
    Positive Years = 93%
    Positive Quarters = 72%
    Positive Months = 70%

Happy to provide any more details that may help understand the strategy better.

Thank you for taking time to provide feedback.

2

Do you know why 2008 was a losing year? Are your signals based indicators that require optimization?

3

Thank you for the question.

2008 was a losing year with -14%. To put the -14% in context, the portfolio gained over 150% before that. The reason it was a loosing year was because the system tries to get back into the market with small bets and given the prolonged nature of 2008 crash system tried a few times to get back in and lost money. Please see the charts and the calendar returns at the link provided above.

The system has to maintain balance between how long it sits out before getting back in which means lower returns. A conservative version of this strategy does not lose money in 2008 but obviously the annual returns are lower compared to this aggressive version.

2008 drawdown could have been easily avoided by introducing additional knobs and make the back test look good. But the system was designed deliberately to avoid using any indicators so that it should work in any market environment. So keep it simple is the key here.

System does not use any indicators except for the price, volatility and the correlation to the rest of the stocks in the portfolio to minimize draw downs.

There is no optimization involved as there are no indicators. System has been tested with everything random which includes restricting the selection from a subset of stocks, number of stocks held, duration held and the day of rebalancing. In all scenarios system always achieved better results than the market.

I will publish additional charts shortly to show Montecarlo simulations and other test results.

1 Like
4

Most of the time, back testing doesn’t work with real time trading, I would say about 90% fail in the long run.

2 Likes
5

@TT40, You are absolutely right that a high majority of the systems & strategies will fail in the long run when the underlying assumptions are no longer valid or the anomaly being explored becomes public knowledge. Systems such as MA cross overs, Turtle trading in their vanilla form come to mind that are no longer viable. Curve fitted systems is another category.

Also let me acknowledge that even some of the robust systems can stop working for long periods of time testing the patience of the trader to stick to the system.

But the real question is how does one determine if a given system falls in the 90% or the minority 10%.

If there is one factor that has been extensively researched and acknowledged to be persistent for the last 2 centuries across geographies and asset classes is Momentum.

There is some really nice work done by NewFoundResearch on the topic exploring the “TWO CENTURIES OF MOMENTUM”.

CrystalBall model is based on all of the academic research and combines a lot of techniques to:

  1. Identify the right selection of stocks
  2. Dynamically allocate capital based on market conditions
  3. Step aside when conditions are not condusive for positive gains
  4. Cut losers and rebalance to avoid large drawdowns
  5. Looks to minimize correlations in the protfolio to reduce volatility.
  6. Keeps track of, if the system returns are well within the accepatble ranges to identify when it stops working.
    etc etc.

As stated in the above reply there are no magical indicators or a million rules that curve fit the system.

Logic of CrystalBall is:

  1. Decide whether to be in the market or not based on aggregate market conditions.
  2. If in the market how much capital to allocate based on how volatile the market is
  3. Apply a ranking algorithm based on price to decide on the candidates
  4. Rebalance at a certain frequency.

I am in no way saying that CrystalBall is fool proof but, rather that it is built based on proven academic research principles with a priority to preserve capital and gradually build wealth.

Only time will tell what the future holds, but I have been monitoring Crystal ball model for the last 15 months for the signs to see if there are any cracks.

Even after all of that any system can fail as no one can predict the future and trading involves many risks.

Thanks for your comment.

1 Like
6

Strategy not visible
This is a “test” strategy and is not visible to the public.

In the meantime, perhaps you would like to explore these other cool Collective2 strategies instead?

7

Probably want to rephrase that. “Large losses prevented by small position sizing, etc”.

Large cap stocks can tank just as easily as small caps.

Never met a backtest on here I didn’t like :hugs:

2 Likes
8

Lol. Yes. that did not come out right. (Note to self: never post long messages in the middle of the night). What I meant was exactly what you said. Thanks.

9

Folks. I just upgraded to a pay plan and lost access to the strategy I created over the weekend. Did not realize that I would lose access.

So I recreated the strategy with the same name and switched out the link to point to the new strategy in the first post of this thread.

Since I recreated the signals at the market price just now, the overall profit % for this month will be lower slightly than what it should be but in the grand scheme of things it shouldn’t matter as much.

Thanks for all the comments so far.

1 Like
10

agreed… back testing means almost nothing, 2017 was one direction bull, 2018 is choppy as hell, markets are ever changing

1 Like
11

We are up against some of the biggest quant funds in the world, Renaissance Technologies, the $58 billion quantitative investment firm founded by former military code-breaker Jim Simons, has like 100 PhD writing algo.

The better coder wins.

12

@PatrickLynn,

With all due respect I have to differ with that statement slightly.

In the context of curve fitted systems, and when there are a ton of them all around, such skepticism totally makes sense.

But, If a system is designed properly, with minimal rules and can be tested across multiple market regimes, without performance degradation, that should be the quickest way to assess if it is
worthy of trading or not.

Take the alternative case, which is a discretionary system. If one were to follow a discretionary trader, without a multi year record one will not be able to draw any conclusions for a long time. There is no shortcut here.

On the contrary, a well designed system can be tested in multiple ways (randomizing everything) and estimating the worst case scenarios very quickly to see if it’s any good as long as the system
developer is open & willing to publish those test results for scrutiny.

One might say, how do you know if the system would work in future. Fair point, but I’d argue that compared to a discretionary trading system, a well designed system may be a better choice, imo, simply because one can run 1000s of simulations very easily to test scenarios than a discretionary system which can’t be tested easily.

And, who is to say that the discretionary trader will continue to perform/trade the same way into future. Life can get in the way and trading style may change or a string of losses may mess up the
psyche of the trader or the markets may change. Question what is the time frame one should wait to assess a discretionary trader?

In my case, to show that a better designed system can be assessed quickly to draw some conclusions, I invite anyone on this forum to provide a random list of 75-200 of largecap stocks (which is what CystalBall trades) and I am happy to publish what the results will look like.

If the system is curve fitted the results would most likely be miserable vs. if it is well designed it should perform similarly, meeting the system objectives. Better results at lower draw down rates.

To me a back test is a useful tool to assess a system quickly, provided the system can be tested in an open way.

Thank you for your comments and I really appreciate the contribution.

2 Likes
13

It is because you know what you’ve done. If you can share details of your backtest including system rules, financial data used for system development and verification, approaches used etc, then after reviewing of all these details one can agree that this particular backtest is appropriate. Or not. Otherwise one should trust you (the noname person from the internet) that your test was done appropriately.

Looking on the amount of successful backtests appeared recently in the internet, and at the same time knowing that no billionaires came from quants community recently, one can really doubt that the backtests are useful. :slight_smile: Thats why unless you share all the details of your backtest, not the only equity curve as usual, the backtests can be considered as useless info and only C2 verified track can be counted as something valuable.

3 Likes
14

dont we all drive in our car only looking at our rear view mirror ?

1 Like
15

@JITF, Point well taken.

The challenge here is to determine where to draw a line in terms of protecting the secret sauce without completely losing the edge. After all we are all trying to make a buck or two :grinning:

But I totally understand what your ask is. I can think of 2 solutions here as a compromise to build confidence on the system without giving away the system completely.

  1. Make the system available for potential subscriber testing so anyone can play around with various time frames over the last X years and allow the user to select random instruments to see the results.

This way, you do not have to trust the no name person :grinning: but see the system behavior for yourself with your choosing of instruments and time frames.

At this point I do not yet have that kind of infrastructure setup but, let me see what I can do.

  1. A quick alternative to the above is to give me a list of your selected instruments and I can run the test and provide the results to see what they look like, which is what I was suggesting in the last post.

I totally understand the skepticism and I am in the same boat too looking at various systems published here on C2.

While it is nice to be a billionaire, my purpose behind this system is to have a ruled based way to generate market beating returns (hopefully) with lower drawdowns because the alternative is to be at the mercy of the market unless you have the skill to pick the right instruments at the right time and get out with profit for a long time.

Agreed. Happy to share any additional details I can behind the equity curve. Somethings I can publish easily are all the trades behind the curve, randomized testing results, additional stats, montecarlo simulations etc. Let me know what else.

In the end though even after having a long history of data on C2, we still need to trust the system developers to some extent to not change anything behind the scenes either intentionally or accidentally when conditions change and that risk needs to be considered in trading any system along with others.

Thank you for your thoughtful comments.

1 Like
16

@OSUTIA, Actually we do this all the time in life if you think about it. With the back test what you are doing is drawing inferences/coming up with some rules based on your past choices to see if the knowledge gained will lead to future successful outcomes.

For example:

  1. When a kid spills liquid 10 out of 10 times, 11th time you know that the probability of a spill is very high.
  2. When you go to a store and have bad experience 5 times, 6th time you are not going to think that you will have a great experience (while it is possible).

The list can go on and on but you get the idea. What we are doing in each of these scenarios is looking at our past experiences to estimate the probability of an outcome. So in that sense you are driving looking at rear view mirror.

To me a back test is very similar. As long as the system is not curve fitted to a time frame or a particular set of instruments, what it provides us with is a way to estimate what the probability may be of a desired outcome based on hundreds or thousands of data points (past experiences in life analogy).

Thanks for your comment.

2 Likes
17

Yes, thats the key word (curve fitted). If your backtest was not curve fitted then you should at least know more or less what to expect.

But as the saying goes…the proof is in the pudding…give it a run and we’ll see how it performs like that first driving lesson…lol

18

ah got it. Please let me know when the next 10% correction will be and also what i should buy on July 9th of 2018. Since we had so many July 9th in the past, dont we know what should happen the next july 9th?

Backtest is a marketing gimmick invented by mutual funds and a sale strategy to overcome objection by their clients. “Ah, look at 2008. We only down 15%. and 2017 we are up 30%. but past performance does not guarantee future results.”

19

@OSUTIA, Apologies, if the statement above came across as having the ability to predict. There is no bigger fool than someone who says they can predict anything in the market.

But what you can do is:

  1. Look at what the market conditions looked like during a certain period of time, whether market is going up or down on a daily, weekly, monthly, quarterly frequency basis and decide if the conditions are right for you to be in the market or not now.

A simple trend following example would be that if 15 out of the 17 stocks in a given industry crossed above their X-Day MA, one would go long the industry because that worked in the past. There is no prediction here but a hope that since 88% of the stocks in the industry are going up, a trend follower would be inclined to take a position. If in the past this logic didn’t work you wouldn’t have any trend following systems on C2.

Similarly, someone who looks at fundamentals, would look at certain periods in the history to compare the current period to decide if a stock/industry/overall market is cheap or expensive to take a position or get out of a position knowing what followed those periods.

Same with the mean reversion systems.

In all these cases, one is looking at what happened in the past during certain periods, and the market conditions at the time, to compare the current market and decide whether to be in the market or not. Be it a trend follower, a fundamental trader or a system trader.

Trading frequency is a choice of the system developer depending on how the system is designed. for me daily data is too noisy to asses the market conditions, hence my choice is monthly evaluation. This is similar to choosing 5-Day MA vs 50-MA vs 200-day MA to estimate a trend.

To be clear on the logic in my case, the system looks at what happened in the last X months and what are the current market conditions and decide on a monthly basis what to do, be in the market or not. This is not a prediction of the future but rather a reaction to what happened.

Thanks.

1 Like
20

@QuantFour

Lets see you proposals:

1 - It is still black box. Just set optimization behind this service/infrastructure and your system will shine bright on almost any timeframe and any instrument. :slight_smile:
2 - Just optimize your strategy on the given instruments and return best results.

So on my opinion it is share all or share nothing. Any partial solution still gives the trade leader means to trick the system/signal buyer. :slight_smile:

Sure, every moment we as subscribers hope that the trade leader will not go crazy. But 1 year of independent C2 track record gives more confidence than 10 years of backtest equity curve provided by trade leader upfront. And 1 year of C2 track record which fit exactly into last year of the 10 years of backtest equity gives even more confidence. :slight_smile:

Again, I am not against backtests, I just know the ways how to make them looks really good without hard efforts.

2 Likes