This is a mobile optimized page that loads fast, if you want to load the real page, click this text.

Tools being used by your trading competition

Jump to new

Greetings --

That article is from 2009.

This field is changing at an astonishingly rapid rate. Consider anything older than two or three years as dated -- perhaps still valid, but needing review.

Best, Howard
 

Greetings --

There will not be any four line examples, such as are possible in AmiBroker or TradeStation.

Start with the material, including examples, I have published. See pages 15 through 447 of "Quantitative Technical Analysis."

Best, Howard
 
Why not? So we have to buy your book to see any proof or evidence? Surely one of the pro Quant and ML guys can show us the end result of what you're ramping, no need to show the code behind it. Just the predictions or patterns it finds that makes it silly to not try, what you've been arguing about all thread so far.
 

Howard's work is as solid as it gets, from the scientific perspective. Remember, the scientific perspective is not the only perspective. Those with great feel for the markets can do just as well, perhaps better, imo.

This is the one I worked on earlier, showing in- and out-of-sample testing. DAX, daily, includes brokerage, 5 contracts. You can see how it breaks down and becomes untradable when it runs on unseen data (after the yellow line). Still working on it.

Unfortunately, the sudden big drops that ruin the OOS curve come without warning, so dynamic position sizing may not help me here. Howard, if you're reading, is that correct? They seem unheralded.

 
Last edited:
howardbandy said:
Greetings --

That article is from 2009.

This field is changing at an astonishingly rapid rate. Consider anything older than two or three years as dated -- perhaps still valid, but needing review.

Best, Howard
Click to expand...

It makes it more valid that even more competition has gotten their hands on same resources and thus even more crowding out ?

"Quantitative Trading Systems was originally published in 2007, with the revised Second Edition published in 2011."

Time for a review ?
 

I'm not doubting his work(nor am I doubting tech's trading ability), but I'm yet to see any of his actual trading or systems that he's using live(if any at all), or if he just writes books and does webinars. The point is that he and tech have been going on for 11 pages now arguing back and forth stating how good the Quant skillset is and how helpful it can be to your trading, so how about enough of the talk and show us, without us having to buy anything, what are they doing related to quantitative stuff that is helping their trading so much so that they're recommending everyone at least give it a go, otherwise it's no different than someone coming on ASF and pumping up his "Make 3000% in 1 month" strategy that when someone asks for the "Let's see then" all goes quiet or he diverts their attention to something they have to buy first.

It's really quite simple, as I mentioned earlier, they don't need to disclose any special little secrets or formulas, but being pro-quant guys I'm sure they have stats on how their strategies are going, if they're actually running any, which they could easily show us, or how it's helped them to find an edge or whatever. Surely there's something, we're up to 11 pages of back and forth arguments with no evidence or stats backing it up.

I personally think machine learning looks promising myself, but I don't know enough about it or haven't coded anything up myself yet so I can't show you how good or bad it is, whereas Howard clearly seems to know what he's talking about and has being doing systems for years and tech is working on some mystery Quant assisting tool, so they're both more advanced than I am, and one way to prove your point in an argument is to show evidence/proof of what you're sprouting is legit. Then we can all be amazed and our eyes opened and can start talking about the useful things
 
Sam

I cant agree more.
Hopefully our private mails have helped you with what I'm involved in.
 

Greetings --

Yes. And there is one.

"Quantitative Technical Analysis" is the update. It omits quite a lot of material that is based on "traditional" technical analysis, as well as material that I no longer recommend people spend any time with. It also focuses more on the scientific method that essentially all research and development (somehow except the trading system community) embraces.

There are several original techniques in the QTA book; and incorporation of several that were introduced in earlier books such as Modeling Trading System Performance. The result is a clear description of the "stack" of processes required to develop and validate trading systems, and to manage trading systems.

The QTA book contains clear and detailed explanation of the process of translation from traditional trading system development platform to a Python-based trading system platform and then to machine learning.

A lot of the material is available at no cost -- on the book's website, in forum discussions such as this one, and in the presentations I have made and posted to YouTube.

For trading system developers who want to stay with AmiBroker, Quantitative Trading Systems continues to be a useful book. It is a gentler introduction than QTA. I was planning to take the QTS book out of publication, but several people said they are using it in their classes and they want it to be available. It continues to steadily sell a few copies every month.

-------------

Yes, the markets are becoming more efficient. There is less profit available. To any traders, but to retail traders using end-of-day data in particular. Competition is definitely increasing. There is definitely crowding out. Testing trading systems over periods of time show decreasing profit and increasing efficiency. All to my point to be very careful to analyze any trading system you are using or planning to use with respect to its risk, and be quick to reduce position size when any system enters a drawdown.

I hear the people who think and say that people are better at identifying profitable trades than machines. The evidence is otherwise. In every judgement-based task, well designed algorithms are better than human judgement. Perhaps I need to say nearly every. But that is changing rapidly, as algorithms are demonstrably better today than they were yesterday. Algorithm superiority is clear in board games such as chess and recently go. Also in medical diagnosis, identification of irregularities in medical charts and images, legal research, engineering design, etc.

And in identification of entry and exit points for trades. Some members of this group may be exceptions. But most are not. I recommend that everyone read Dr. Daniel Kahneman's book, "Thinking, Fast and Slow" for a thorough discussion of the many ways we fool ourselves. Dr. Kahneman was awarded a Nobel Prize for his work in this field.

Best regards, Howard
 

Yes I agree with trading systems showing decreasing profit over time. Set rules are optimized on the past and the market will always seek to break them.

I don't see the parallel between chess/go & trading. In board games you have a LIMITED number of outcomes. If your opponent moves here, there is only an X number of spots to move and outcomes to be played out. The machine will calculate those and every single possible outcomes that can be played out and beat the human. This is not true in price - there is no limit to the number of outcomes of price, it is not bound by rules of the board game that you have to put the pieces on designated spots. Poker would be a closer parallel to the markets and as Omega has posted before the human were beating the machines.
 

Greetings --

Everything should be free? Authors and researchers should not be compensated? Trade secrets should be openly discussed?

Pardon me while I resist a rant about the inequality of compensation.

Much of the material in my "for sale" books is available free -- on each book's website, in the forum threads, in the presentations I have made and posted to YouTube.

I buy, read, and study more than one hundred books every year. I also read articles posted to website, watch YouTube videos, subscribe to Coursera courses, correspond with colleagues. Some is free, some has a monetary cost. Some is worthwhile, some not. Part of being competent in a professional is continuing education. Even if the materials are free, the time is not. Books are inexpensive.

---------------

To your question about the patterns being found. One of my points is that machine learning models often do not have the ease of interpretation of decision tree models produced by AmiBroker or TradeStation. The model is, in general, the coefficient matrix of the solution to a large set of simultaneous equations.

------------------

At the risk of making things muddier rather than clearer, here are some of the graphs that represent a trading system that were produced during development.

Trades produced by a trading system form a distribution. The image below is an illustration of the distribution of maximum drawdown of a machine learning system when the position size has been set so that the drawdown remains within the trader's tolerance. That position size is "safe-f." The terminology is that position size is such that the results are "risk-normalized."



The image below is the distribution of terminal equity when that system is traded at safe-f.

Any single equity curve posted would be just one data point from the distribution. One possible equity curve that might occur if the system is traded in the future. Several equally likely equity curves are shown in the chart that follows. We have no idea which of them it will be, if any.



The decision whether this system is tradable or not depends on the alternatives available to the trader. CAR25 of this system is 11.1%. If he or she has an alternative that has higher risk-normalized CAR25, then that alternative should be traded.

Best regards, Howard
 
ThingyMajiggy said:
Surely there's something, we're up to 11 pages of back and forth arguments with no evidence or stats backing it up.
Click to expand...
Another one of those donkey chases the carrot games. Realistically most people are not going to learn computer language because of the time and brains required. For those that do or have, maybe they will show up one day with a positive OOS equity curve and some guidance on what data they used to get it. The meat in the sandwich so to speak otherwise it will be this persistent dangling carrot theme.
 

To make this a fair request I offer my eyes and brains to be pitted against the machine. Will be available in a few weeks to trade against any retail machine system in a live forward going setting. I don't like my odds against the 80% win rate 1:1 RR mentioned earlier but I am happy to be humbled for fun and education. PM me if interested and we can set something up.
 

No everything should not be free and the authors and researchers should be compensated, I've already mentioned no trade secrets need to be released. But what you're saying is you can't test drive the car until you pay for it.

Everyone here involved knows you don't need to reveal anything to do with the actual strategy or trading method you use, all you need is an equity curve, a % return over X period, drawdown etc. stats, you know the deal so saying all the above is just a bit silly as nothing needs to be revealed that ACTUALLY makes the profit. Just the results of it.

To me, I'm more than happy to pay for someone's service/books/software/guidance, but first I like to see evidence that what I'm entering into is legit. That just makes sense to me, proof, then you can have my money, simple. Not the other way round.

Oh and it shuts everyone else up and we don't have to have this 11 page debate about what is better

Oh and it's kind as you've already mentioned you don't need more money.....but that's no incentive is it lol, that's just the resolutions everyone makes each new year.
 
Enjoying this exchange.

For stationary systems, quant will eventually win hands down. Stationary means that the system itself basically remains unchanged. Laws of physics don't change and so mechanical structures outside of the quantum realm are basically stable. If you are training an algo to play chess, the rules never change. The algo will eventually win (and has). Medical diagnosis is mostly like this. The learning rate of an algo designed to spot anomalies in a scan can learn from more scans that could be viewed in a lifetime. Eventually, sort of like chess, the computer becomes better than the radiologist on their own, and possibly better outright. This day will come. The target never changes, the competition never evolves. The computer just gets better and eventually, if the problem is capable of codification, the algo is the best thing at it.

When you put yourself into situations where the competition evolves...it gets trickier. Unless that evolution is sufficiently predictable (ie. I need to know how you will react to something I have not shown you yet), it gets a bit shaky. Situations involving humans who adapt and algos which are also seeking to adapt moves things into a primarily non-stationary situation. Generally, both machines and algos suck although the efforts can be valiant.

Sometimes, systems are chaotic. Like weather. So it is not really stationary, but enough to allow adequate prediction. Weather forecasting is a great example of applying quant in difficult and unstable situations. But weather is not attempting to beat the forecasters.

Whilst driving involves siutations where objects react to one another, they do so in a way that allows driverless cars to navigate around more safely than human ones. On average, I would hands down rather be driven by a Google car than by just about anyone on the planet. So this is a non-stationary system which displays enough order for quant to definitively exceed human capability.

---

Now, a market is not a uniform mass. There are pockets in it which are all of the above. So, whatever your view, you can be right if you qualify it enough.

---

Quant in equities generally applies a very different approach to outperformance than via traditional fundamental analysis. The edges are not talked about in terms of stocks, but by factors. If you think FMG has a 60% chance of break-out or NVT is super-cheap for a five year play, the equivalent for quant is to say something like my low-vol factor will produce a favourable outcome. Any individual stock with a decent exposure to that factor would have a very low edge compared to the FMG/NVT examples. But much more diverse portfolios are built to offset this. The edges look very different. We can talk about satellite imagery across crop fields etc... but let's not.

In a situation like the market, which is constantly evolving, there is an opportunity for someone who learns well to assess situations faster than an algo could. This can happen in situations which look kind of similar across different companies, but where data may not be readily comparable from a data analysis sense. With training, it is perfectly plausible for a person of talent to see situations which rhyme at sporadic intervals. You might spot similar kinds of market developments in different industries, for example, and successfully template one on another. An algo has a very hard time knowing what to look for in these situations. If this is your gig, and you are actually good at it, all power to you. If you are mining price series with nothing else to add to that, it is a very valid question to ask why the vast tracts of algos run by highly profit motivated people haven't found it already. They might have, but do not have sufficient capacity to fully exploit it or suffer other structural impediments.

There has been a revolution in data science in recent years. It is truly amazing. Tools from all sorts of fields are applied to others, including the markets. It might well be that this will seem more like applying astrology to markets. I doubt it. I think it will be more like the benefits I gain from using a computer today whereas it as an HP 12C when I first started work. It was an impressive kit and the maths is the same. I can just get a heck of a lot more done now, as can anyone else who does the same types of calculations. It's still me writing this stuff.

---

Who are you taking money from? A machine? Another human? A cyborg? If you don't know, buy an ETF or understand that this is a speculative or otherwise entertainment related activity. No problem with that either.



Have a good day.
 
Last edited:

Hi GB, and all --

As we try multiple approaches to a trading system, we see many test results. Typically, it is an exceptionally nice equity curve that catches our eye. A single equity curve is a single data point drawn from the distribution of trades. The one we stop to look at is invariably better than most that could be drawn from that distribution. So be prepared for risk-normalized results to be lower than the single equity curve would suggest.

The analysis goes as follows:
1. Any set of trades can be analyzed.
2. If you are using a rule-based model and want to estimate future risk and profit potential, use only those trades that are out-of-sample and selected by an objective-function-directed search.
3. If it is present, remove the effect of position sizing. At this point, we want the percentage (or points or dollars for futures or FOREX) gained for a constant trade size. Sort this list.
4. State your personal risk tolerance. For example, "I am trading a $100,000 account, forecasting two years into the future. I want to hold the risk of drawdowns greater than 20% to a chance of 5% or less."
5. Work through the procedure of computing the cumulative distribution function of drawdown for two year trade sequences drawn from that list. This procedure is explained in detail in several of my presentations, posted to YouTube, and free. It is also in two of my books -- one using Excel code and the other Python code.
6. Adjust the position size so that the risk at the 95th percentile is 20%. This is safe-f -- the position size that will maximize profit growth while holding drawdown to within your tolerance.
7. Using safe-f, compute the cumulative distribution of terminal equity at the end of two years. Look at the 25th percentile. Convert the percentage to compound annual rate of return. This is CAR25.
8. Compare CAR25 from this set of trades with CAR25 from whatever other alternative use of the funds you have available. Pick the best one.

To maximize equity growth over a period of time, we want:
1. A high expectation. High percentage gained per trade. Express this as geometric gain per trade, where a gain of 2% is expressed as 1.02. Call it "g"
2. Many trades. Say there are "n" trades in the period. Terminal equity for any system is determined by only two numbers: n and g. TWR = g ^ n. Final equity as a ratio to initial equity is expectation raised to the power of the number of trades.
The TWR calculation holds for all sets of trades. 50 1% trades are always better than 1 50% trade, due to the compounding effect. 1.50^1 == 1.50 1.01^50==1.64 But that says nothing about drawdown.
3. Limit the number of losing trades and limit large losing trades. Either increases the likelihood that there will be a drawdown that exceeds your tolerance.

Together, these form the trading mantra that maximizes return while minimizing risk:
Trade frequently
Trade profitably
Hold a short period
Avoid serious losses

The first three are easy. What to do about losses?
If the problem is many small losses, look for a model that is more accurate. Again, this is pretty easy. Any fast oscillating indicator based on price will show oversold accurately. Try RSI(n), where n is in the range of 1.5 to 3.0. Or DPO with a fast adaptive moving average.
If the problem is a few large losses, look for a model that avoids them. This is much trickier and often requires adding rules to the model. If the model is being built using machine learning, try a two-step model.
Step A -- Avoid toxic trades. Train the model to identify those conditions that result in the large losses.
Step B -- Proceed as usual after first blocking -- forcing to be flat -- all those days predicted to be toxic.

If the CAR25 of the resulting model is adequate, the system is passed to trading. There, dynamic position sizing will control the trade-by-trade position size so avoid serious loss.

--------------------

Specifically to the equity curve posted. It appears that there are several large losing trades. You are correct that dynamic position sizing will have difficulty adjusting quickly enough to keep a system like this out of trouble. The model needs to be improved.

I hope this helps, Best, Howard
 

I am not drawing an exact comparison between chess and trading. The examples are to indicate the skills that are available using algorithms. Skills that were thought to be unique to humans until very recently.

In simple games, the computer can compute all alternatives and choose the best. In complex games, such as chess and go, the number of paths is too large to be enumerated and exhaustively analyzed. Rather, algorithms assess probabilities based on current conditions -- just as do people.

You are correct that there is a great deal of variability in tomorrow's price. Your approach to trading is probably different than mine. I do not need to know tomorrow's specific price. Up or down is sufficient for many models. Greater than or less than is sufficient for others.

Depending on which game, which research, and which tournament is reported, humans do not do well against algorithms in poker. But that aside, humans are not the primary opponents in trading -- algorithms are. It is not me against a person at another terminal. It is my system against all other systems, including those of Goldman Sachs and James Simons. The gains are distributed among the best systems, taking the amount gained from the worst systems.

Best, Howard
 

Hi Minwa --

Let me understand. You are willing to be the counterparty for any trades I want to make for two weeks?

If I Buy ES or SPY at today's close, you will Sell. My gain (or loss) is your loss (or gain)?

My trades are typically:
A. Market-on-close issued within a few minutes of the close. They may hold for a day, or they may be closed out in the after hours session at profit limits. They may be for futures, ETFs, or options on either.
B. Limit orders to be filled intra-day at what my algorithms estimate to be extreme prices, with exits MOC or limit.

I will pass your offer. But thank you very much.


Best, Howard
 
Greetings --

Some of this thread reminds me of some students I have had in my classes. Every semester or so a young man (they are always male) approaches me with a scowl on his face, shows me the C, D, or F he received on a recent exam or assignment, and reprimands me for failure to recognize his ability. I point out that he is on a path to failing the course. He makes it very clear that, no, he did not read the book, and no, he did not do the homework, but he knows the material and deserves a gentleman's C so he can check off this box and get on with his life.

There is an implicit reading assignment that goes along with my comments. People who want to improve their skills in developing profitable trading systems need to be aware of the foundations of modeling and simulation, computer science, machine learning, statistical analysis, and everything else involved.

There is no gentlemen's C in trading.

I think there is a gross misunderstanding of the purpose I had in mind when I began this thread. I intended it to be strictly educational. To explain what is happening among firms that are using algorithms, and what individual traders should take into consideration.

The amount of denial in the postings is much higher than I expected. Do the math.

Use it, or leave it, as you wish.

Perhaps the thread has reached its limits.

Best regards, Howard
 

Why is it professors always think they're right and theres no other way, when actually challeneged backs away and uses big words instead in big posts full of words. Wasnt it people like this that made the world financial state like it is today? "My ways are right and I refuse to listen to any outside sources. In fact I'll claim them to be in denial or jealous. Oh and what I say goes, dont even think of checking any hard facts to back me up. I'm a professor you know."

A piece of paper saying you're qualified in something doesnt mean what you say is 100% and thats that. If a student tried to do something differently a professor is likely to shut him down because its not "meant" to be done that way. But the way we've "always done it" hasn't landed us in a very good position has it. A student isnt lacking ability because you think he should have done it your way.

People seem to really struggle in accepting that there are multiple ways, and that what they're complaining about and saying about others is precisely what they're doing themselves with their own bias.
 
Sam/Klogg (like??)

Really?

The amount of help given to people in this community by Howard has been enormous. To treat an expert in a field such as Howard with such un restrained contempt when your not in a position to do so is childish.
Denigration to personal attack demonstrates a lack of comprehension and understanding.

I suppose Charles Teo shouldn't be listened to when he advocates using hand held mobiles are your biggest risk for Brain Cancer. Experts earn their expertise. In which area is yours?

Grow up.

howardbandy said:
Perhaps the thread has reached its limits.
Click to expand...

It has.
 
You must log in or register to reply here.

Similar threads

Getting Started in Machine Learning for Trading
2 3
Replies
50
Views
22K
entropy
E
Python - getting started
2 3
Replies
43
Views
19K
fiftyeight
The Beginner Trader's Manual Guide
Replies
9
Views
20K
commoyo
M
  • Poll
Need help! Deciding what to study to become a trader!
2
Replies
20
Views
9K
artist
Menu
Log in
Register
Cookies are required to use this site. You must accept them to continue using the site. Learn more...