I do very well with the Strategy Analyzer. I get the same results as the Strategy Analyzer when I manually track the trades. So, I have confidence it does what it's supposed to do. However, you MUST know EXACTLY how your bar of choice is built. That is the most important part imo. I've seen too many people who do not understand how their bar is built and wonder why their backtesting results don't play out the same way in live/sim trading. Then when told of their error, many still continue their fruitless efforts and complain that back testing in NT and the strategy analyzer are worthless. Bottom line is you can't just accept what the Strategy Analyzer reports without verifying in excruciating detail that your strategy is taking and managing trades correctly.
The following user says Thank You to MWinfrey for this post:
Mike - thks and appreciate your experience.
I use basic Tick charts, so using like 20 ticks, 40 Ticks for backtesting.
not exotic Renko or other charts . is Tick chart ok? can I ask what bar type you use?
or what bar type preferred for backtesting. when i dont have problem in changing to bar type
The following user says Thank You to emini_Holy_Grail for this post:
Tick and volume are ok. However, if you are trading patterns, they have a tendency to change when comparing a live chart to the same period based on historic data. That has nothing to do with NT though. That's just the way tick and volume are. I've never been able to get tick and volume to provide a consistent view of patterns from day to day or person to person.
I use BetterRenko for everything because I get a consistent view of the market and know that a pattern that i see today will be there when I view the same pattern when backtesting.
The following 2 users say Thank You to MWinfrey for this post:
Mr Winfrey, May I ask how a person who does not know coding is supposed to do that? Also why bother with a backtesting software if you cant trust it unless your system is very very complex, no?Id like to hear from others who have backtested systems and found they worked in the backtest but do not in forward test or live trading?
I myself bought a neural network software that allowed me to tweak it from "conservative to aggresive and I never knew how it got its results. I didnt care. All I knew was that it showed me stocks that produced over 100% annual returns with over 60-70% winners. Thats all I care about. Every stock I then papertrader AND real traded completely fell apart. This was on daily charts only by the way. So this begs the question......"Is every system destined to fail at some point?" I ask you all before you even think of answering, to look at the turtles system from market wizards, the most successful system ever devised. It now sells on the internet for like $1500. Why? It doesnt work well anymore!!!!!! So if the simplest best system failed(luckily those guys made a fortune before it failed) why shouldnt anything else?
@Daytrader50. good question about being a coder. Quite honestly I don't know how you would do the things I mentioned without being a coder. I selected NT several years ago when it was version 6.5 because I could write my own stuff and be able to test things that other people were saying they use for trading. In fact I was specifically wanting to back test Woodie's CCI patterns. I won't go in to the details of that effort except to say I found out very quickly how unprofitable they are. Prior to NT I could only collect data manually and put the data in a spreadsheet. That's a horrible existence. However, with NT I found that my coding skills were limited but was able to work through all that learning curve of learning to program. I have some programming experience in previous lives so I wasn't starting from scratch. Regardless I kept at it. All the negative things people talk about with backtesting are pretty accurate and I've experienced them all and I mean that literally. However, I kept at it keeping in mind all the bad things people were saying and experiencing some bad things on my own. I dealt with each issue individually attempting to minimize their effect on my back testing results because my goal was to get the same results in live/sim trading as I do in back testing. I was relatively successful in overcoming those bad things that happen by using time based, range, and BetterRenko. Was never successful using volume or tick. That's basically my experience in a nutshell. Now back to your question of "how a person who does not know coding is supposed to do that?". If you have an aptitude toward programming then dive in and learn it. But, I'm not so sure that's the best way to go. For me I believe it was. Probably delayed becoming a successful trader.
As far as complex systems, my systems aren't complex at all and I have come to trust NT's Strategy Analyzer. I know it's quirks pretty well. I don't know all of them I'm sure but I think I make up for not knowing everything by questioning everything I see and addressing each issue as they arise. Also, I each live/sim trade with that same trade based on historic data. If there is a difference I find out why.
The following 5 users say Thank You to MWinfrey for this post:
1. Problems with backtest engines (which is one cause of problem 2)
2. Backtest results not agreeing with Live Results
Here is my quick take:
1. Problems With Backtest Engines Every backtest engine has limitations. Mr Winfrey hit that 100% correct. I use Tradestation, and after a few years of using the software, I came to realize that there are certain assumptions a backtest engine has to make (about fills, or the path of price during a bar, etc) that make the results suspect. So, for me the solution was to primarily use market orders only (with some stop orders), and avoid limit orders. I also avoid scalping type strategies. The end result: I can get pretty good agreement between backtest and live results.
2. Backtest results not agreeing with Live Results Eliminating issues with the backtest engine is just step one. Even then, it is really easy to overoptimize, curvefit, using hindsight bias, etc. - a million different things you can do to render a backtest invalid (by that I mean it will not produce comparable results going forward). For example, did you use out of sample data to verify your test? Did you use walkfiorward testing? Did you incubate it - where you put the strategy on the shelf for a few months, then come back and evaluate how it did during that time? If you don't do those things, your chances of success go way down.
Unfortunately, information of how to do all this is rather scant, or at least it was when I learned it primarily on my own, and primarily by losing money and wondering "why don't live results match real money results?"
1. If your backtest results seem too good to be true, it probably is too good to be true.
2. Avoid scalping strategies that may be "gaming" the backtest engine
3. Keep your strategies as simple as possible - then it is easier to diagnose issues
4. Use out of sample and/or walkforward testing
5. Incubate your strategy
6. When first trading real money, keep it super small until you gain confidence
Last edited by kevinkdog; November 1st, 2012 at 08:12 AM.
The following 6 users say Thank You to kevinkdog for this post: