Welcome to NexusFi: the best trading community on the planet, with over 150,000 members Sign Up Now for Free
Genuine reviews from real traders, not fake reviews from stealth vendors
Quality education from leading professional traders
We are a friendly, helpful, and positive community
We do not tolerate rude behavior, trolling, or vendors advertising in posts
We are here to help, just let us know what you need
You'll need to register in order to view the content of the threads and start contributing to our community. It's free for basic access, or support us by becoming an Elite Member -- see if you qualify for a discount below.
-- Big Mike, Site Administrator
(If you already have an account, login at the top of the page)
I currently run almost 300 strategies live on FX ( small accounts) some for over a year even thought they trade very small amount I still have to have 25k locked to trade them all. Equity wise I am on plus. They are combined in portfolios about 20 of them. During this year I tried many things. I copied the 15 best strategies out of the oldest portfolios and made a new one and thought (he he dude) you will be rich now. Well bullshit this new portfolio went straight down and surprisingly the Original ones EQ curves started to really perform well even thought they still included the 15 strategies. What does this tell me. Well diversification is the only thing that matters. It is hard to diversify in FX market and to risky in futures for my capital base at the moment but you can using different type of strategies in FX, timeframes and you can avoid having to WFA by having correlated strategies on the same instrument working with different set of exits and targets making them together handle different types of markets and you can easily size them to even out the EQ curves . Watching these 300 strategies trade is fun they take on average 26 trades per day and I want to add more. My plan is to trade the equity curves by copy them from master accounts via an WEB/php/MSQL server and control which ones will be allowed to take a trade or add size or lower size based on whatever I can up with which can be some standard deviations from mean, drawdown value, consecutive looser or winner or other indicators. Currently I am struggling with getting the system up and running. When it works If ever I will share the experience . I might add that the 15 strategies did have 40 % drawdown directly after I moved them out but today they are today up 18 % after 7 months.
I have been trying to think about what any real lesson might be gained from this. And, I think a few things stand out in mind. The first is market cognition. One of the chief problems with trading systems is that without expert trading insight then such systems are more likely, due to convenience bias and significant difficulty -- not impossibility -- to develop around such problems, to be built around the less important factors.
One way to think about it is that a great discretionary trader will identify the factors that are influencing their market and they will then capitalize on those factors. Sometimes it is impossible to prove that these were the factors influencing the market until after the fact. In other cases, systems might be able to be built around such factors successfully. What is relevant is that the market cognition is much higher and more likely to be sound. On the other hand, trading systems are more likely to try to find the I'm imagining "NASH-like" equilibrium for a mix of factors under the assumption that they are random. What I think is relevant is that if a system was tested and did well on out of sample performance that the causes of the failures are probably not due to simply that the system didn't work in the past but rather due to several specific types of causes. One problem with trying to develop systems in this way is that certain persistent effects can drown out other types of effects. For example, a system developed on a market that is mostly mean reverting but has some good momentum trading times will be unlikely to discover those momentum effects esp if they only occur as a result of non price data.
Sometimes having a good insight into market dynamics combined with exceptional technical read can provide superior benefit. A good example is that markets tend to be more mean reverting during periods of uncertainty. A good discretionary trader can pick up based on the news, sentiment, etc. regarding the uncertainty. This will give such a trader insight into the possibility that the market is mean reverting combined with ability to see if that is what is happening. Is it technically possible to develop this into a system? Sure but you will need special data feeds or you will have to result to price proxy data. In other words, system developers are more likely to be led down the wrong path because of what is available in most system development packages. Imagine, also, for something like Bitcoin -- where you have some future event that is known that can influence the price. Can you develop a system that takes into account future impacts? Sure. But most will not because it is difficult.
But, let's go back to why systems that worked can fail. A system that has many trades might be prone to small changes in profitability. For example, if the market is 52% more trending then mean reverting over some sample period then a small shift can cause impact. Changes in volatility are the most difficult for both discretionary and systems.
There are several reasons: the factors driving the market can change, the mix of the factors can change, the expectation of the patterns can change in a significant or small way, the frequency of the patterns/trades can change, etc. The good news is that it is possible with serious investigation to probably identify the causes for any particular system. The bad news is that it is not clear that identifying the differences will clearly state whether the system will work in the future. I can imagine, for example, if a market like Bitcoin is getting a lot of news stories, a lot of press, then the technical patterns might be different then if the market isn't getting a lot of attention. Sure, it is simple in concept: a classifier for technical patterns based on frequency of news. That is not going to be easy to build or test though. So, you know that most systems are going to be built exclusively on price data. The cognition though factors in the news.
The other lesson is know your enemy. Futures markets are decidedly unfair due to the fact that large liquidity providers can basically take the fair bet against the other traders. It is mathematically given that the more trades one places for a given level of return will reduce the risk needed to be taken on any one trade (provided one isn't trying to profit from outliers-- and even possibly then). But, traders are often told to be selective. Don't over trade. But, if you're a trader: the more trades you take then the more you can make. Selectivity introduces several forms of risk but one of them is the need to bet more per trade and to trade during more risky periods.
The CEO of Virtu, a market making firm with multiple stretches of multiple years without losing days, said "we don't trade quantitative models and that's why our models don't break." Basically, their model is to capture the spread. If you look at most top tier futures firms: they are all market makers.
Let us take a moment to notice something important: most retail system developers cannot develop systems that effectively capture the spread or even half of the spread. If you run a managed system then the costs are higher and it is even worse. Of course, you might counter but the spread provides liquidity. But why does retailer need such liquidity? They don't. It is only the whales. And the futures market is designed for the large liquidity providers that can trade for free and the large traders who tend to hold their trades for longer. If you have a (1) larger edge and (2) longer holding time then the spread, via the liquidity, can be a benefit. But, it is based on all the evidence much more difficult to build such systems. I think the large futures spread systematically hurts the small speculator who wants to trade frequently to obtain HFT like returns. Also, when you factor in that futures exposes the short-term trader to valuation change risk then it is even worse. Why do valuation changes hurt the small/short-term trader more then the long term trader? Because, a small trader will typically have both stop and target in market. The target is never blown but the stop can be blown. A large trader might not even have a target in market. Before HFT, there might have been some order to it all: the liquidity provider takes the risk of getting hit but now it is only the small trader who is disadvantages. The liquidity provider is more likely to pull their orders when times are negative and then to stuff the books when times are good. That's why I'd like to see a true 50/50 risk limited futures market for small traders who want to take the fair bet.
Regarding limit orders, some of the more interesting work I've seen on the matter has been written by Dr. Jonathan Kinley (C2 vendor/trader).
The takeaways on my mind are:
1. As a system developer, you must make special effort to avoid convenience bias. You might be well served by trading at least the open or for 2 hours using discretion, even if on the sim, to keep market cognition sharp.
2. While going to market is more sound for the trader who has the larger edge I.e going to market, trading costs cannot be ignored. Even the quantitative/system trader should seek to minimize their trading costs.
3. More trades is better then less trades. You can get in more trades by trading more markets or by trading more often. If you do (1) then you can probably get by with current generation tools. However, the downside is you will need to take a lot of risk per trade relatively speaking. If you try to route (2) then you will probably have to develop your own tools and do the hard work to figure out why other traders cannot make limit order systems work, fix it, and make it work.
4. It is very possible that quantitative trading for a trader with already high market cognition and the ability to do the hard work could provide very superior returns to discretionary trading alone.
Lately the market i trade on have changed alot you can see that from the performance of some of my older portfolios that have been stable but now are turning straight down. I do not think there is any strategies or ever will be any strategies that always work. You will always have to make new ones or update the old ones. The problem is to know when to do it i wish there were an autoWFA feature build in that you can set parameters for when to perform the update automatically. That would be cool.
It has been going down since August the only thing that kept it hanging until september was a great gold trade that finally reverst and gave most of it back.
Belowe is essetially the same type of strategies but they have other settings and have been working great during the same period.
Even though trading many strategies seems to remove the drawdown it definately also flattens the gain it might even become a 0 sum game. But hopefully if you always are in the market with so many strategies you can improve the gain by analysing when to trade them. But for that You need alot of history of live trades. By the way this is all intraday strategies 15M,30M, 60M.
I want to point out something, also. Even if a developer makes successful systems and uses whatever method they use to validate them such as WFA. It doesn't demonstrate that the systems work because the validation method. The developer must keep the the systems that were promising but failed validation and track how they perform. Even this is not going to be conclusive because if the developer has a belief that the systems that pass the validation are better then they are more likely to spend more time to improve and work on them.
If the validation method doesn't demonstrate the validity of a system then what does? There are three somewhat similar explanations. The first is market cognition. It means understanding how markets work, which markets are likely to see action, etc. The second is that it is a combination of market cognition and active tracking and management. The third is a similar but slightly different concept which is an understanding of the nature of statistical measures that are likely to work or fail in trading. The concept is related to the prior but again slightly different. For example, a developer that avoids optimizing a parameter that is likely to be curve fit is demonstrating a type of "strategy or statistical cognition" or just "experience". Realistically, it is probably a combination of all these working together that explain the performance of any developer.
As an example of how one might apply market cognition toward system selection, one might first start with selection of markets that are likely to see volatility or trend. I do not trade interest rate products but I think interest rates, such as 2 years, are likely to rise and/or see volatility. I'm willing to predict that 2 year rates or the prices thereof will rise. As such, I look for any products trading those products. As well, I think it is possible that any markets/products sensitive to cryptos could either see trend or volatility. So, that suggest the NASDAQ. Based on those ideas, I added a few other requirements such as filter by time in market, profitable most recent 3 months, and live traders. As a disclaimer, I have never attempted to predict system performance.
I do not have any relationship to these vendors. And I have no idea how these systems trade. I would likely want to lock to short only or long only based on what I seen developing and my expectation of trend. I have not studied the trades they make nor reviewed the logic by which they trade.
The basket I came up with though would be: CIRUS RVST 30 T-BOND INTI (time in market is high a concern), IRU T-BOND, Dothraki Gold ST, IRU NASDAQ
I will not be taking any action on this as I develop my own systems/trade ideas. I would need to do additional analysis before I'd be interested in that, certainly. However, it might be fun or entertainment to check back and report on how I did. I have a knack for predicting things. My time horizon is around 90 days. I think I would need a bit more understanding as to the underlying logic to trust trying this. It is possible if the systems are designed wrong and/or do not adapt to the regime shifts then they might move to either significant under performance, as well.
But, I agree regardless this may not be a rewarding answer for someone looking to quantify the future performance. However, you could generalize the approach to tracking a measure of volatility for markets traded and analyzing performance in relation to the volatility.
Yes, if you can anticipate a dircetion of a market and then pick strategies that trade affected instruments you have a greater change of success picking a good strategy then just picking something that looks good.
Simple market change from trending to sideways is the simple answer to my failing portfolio in an correlated market as illustrated below. The change in market behavior is identical to it´s performance by time line.
As promised here is the final PnL recap through Friday. At this point I have switched off most (but not all) of the systems and consider the first phase of this experiment closed. Obviously not happy with losing $5k but in the big picture I think it was a worthwhile experiment, the upside if I was correct could have been considerably higher.
The reconciliation below matches brokerage statements to the ¢ but varies slightly to my previous August PnL analysis due to a difference in license cost accounting. The "Currency" column reflects the small gains and losses incurred by have a USD denominated account but having EUR margined futures in the account - meaning I had an implied currency exposure, long USD short EUR.
As you can see almost 65% of my losses are fees rather than trading loses.
Next Steps
As I said previously I still believe that this data set could be very valuable. It contains almost a 1000 systems that in some cases have several years of real, non backtested, performance history. At this point I have written more R code to extract as much system information as I can ~ and also to calculate all sorts of different metrics on that data. My next step will be to use some machine learning techniques (probably Random Forest's initially, followed by an SVM) to try and use those metrics to identify systems to trade.
I have done a ton of hours looking over which systems and really only found one. I did all the look up and background as well as loookup with NFA and found the developer on Linkined and emailed them, It is a firm. I dont like how little it trades but I cant argue with the P/L. Last few months have been rough as I'm up very little. Now IMO anything put out by this firm is good but that is only my option.
My point is:
There are only a few gems out there and it takes a lot of work to find them. Your own homework will save you many times over because you know the background instead of just what "they" want you to know
However, I do believe iSystems is much better then C2 as iSystems developers can not change their systems once their system is put on the server. I.E. there are zero discretionary systems on iSystems
Trading: Primarily Energy but also a little Equities, Fixed Income, Metals and Crypto.
Frequency: Many times daily
Duration: Never
Posts: 5,049 since Dec 2013
Thanks Given: 4,388
Thanks Received: 10,207
Minor Update
Just doing some month end accounting. As I mentioned I switched off most but not all the systems. Glad to report that the few systems I have left on performed well in late September and my account is now positive again! Yay!
On a different note, I've performed my initial Machine Learning (Random Forest) analysis of my data set and must say the results are encouraging. Not excellent but definitely shows some potential. More to follow (eventually).
. I might add that the 15 strategies did have 40 % drawdown directly after I moved them out but today they are today up 18 % after 7 months. 
