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)
@semiopen So you know one platform where they do not use log returns in a list of trades, so everybody here are probably using misleading backtests results. That's quite an non scientific, non objective implication.
I know NinjaTrader somewhat, don't think that is any different design wise... a bell, a whistle there. Log return isn't the only issue.
I think the multiple security issue is the easiest way to grasp what is wrong with the overall discipline.
In futures terms, if you trade ES, NQ, YM... if you have a strategy for one it should really work on all of them - ie. similar returns. I don't really see that with traditional strategies maybe you and everybody else do. If such animals exist why not publish them? If the secret is so precious why not publish one that doesn't do quite as well?
Are you suggesting that you have to analyze every possible commercial offering before deciding to to do something on your own? Guess that's what made this country so great.
The current buy and hold return is LogRet. TotRet is the strategies return. Notice the strategy beats all of the buy and hold and shows a nice average profit of 12 versus a loss of -9 while being in the market a total of 40 out of 93 trade days.
Could you show a strategy that has a similar return on one of these stocks?
Look at all the posts in this thread over 4 pages of small talk.
How is this for the last 400 days? Exactly the same parameters as health care, 191 days in the market out of 400, beats buy and hold by a factor of almost 3. Nothing to see here, right?
So showing a list of trades where an unknown strategy is performing better than buy and hold. What does that prove? It only leads me to believe you have no real experience in algo trading.
So yes if you make these statements like nobody has a clue, and you have the knowledge, write an article, do some research and don't forget to publish the link, so others - maybe not a smart as you - can have look at your reasoning. This approach has lead to a lot of prosperity. It's called science.
PS: also share the link to other promised article about negentropy..
Maybe that is the problem. That is not a list of trades it is a summary of trades. A trade is not a good construction because it just clumps consecutive days holding a position together. These are not consecutive days but a summary of all days together.
Uh no. Futures traders like to specialize in specific instruments, and are looking for behavior specific to their specialty. The strategies I see most people trading are only going to work on that specific instrument. So they're not going to be that interested in log returns since they only care about how it trades on that one instrument.
I understand, my former strategy experience got results that say, sort of worked on IBM but failed badly on HPQ for example. That would be OK if that was the best one could do; that's as good as current software products do.
The examples I posted yesterday all trade on exactly the same signals. When IBM is long for example, HPQ is long, when IBM is flat, HPQ is flat. That's why the days row total is smaller than the total number of days.
rlstreet gave me an idea... I'm writing a sheet with "Long days" and "Flat days" that way I can run securities through that sheet and show comparative returns for buy and hold vs long days on a graph. The long days and flat days are determined at the close of business the day before, so one either buys the security at the close on long days or otherwise goes flat. That is equivalent to what I'm doing now except it is difficult for others to grasp. Have to admit I was getting a little confused myself.
Usually when you develop stuff there is a user you can talk to, in this case, its just myself. It shouldn't have been a surprise that no one else can understand it.
I really appreciate the feedback, sorry that I offended people. FWIW its not really intentional.
BTW: If long beats buy and hold then you can theoretically short on flat days. One of the advantages of the natural log return is that if you know 2 out of 3 log returns, you also know the third one too.
For example, if something has a buy and hold return of 10 and a long strategy return of 15, the other return is 10-15 = -5.
But I couldn't do the the article in the link as it was a followup of another article which was locked. So I had a look at another of your more recent articles. Portfoliostrategy: Peak And Trough Analysis: https://seekingalpha.com/article/4335520-peak-and-trough-analysis
In this article you show how a simple a Moving Average strategy on different lookbacks for 3 sector indexes on objective measurable peak-trough and trough-peak regimes pans out.
The hypothesis isn't very clear in your article but I assume these peak-trough and trough-peak regimes as defined by James A. Kostorhryz is the main focus of this article. I presume you want to prove somehow that these regimes gives investors/traders unique opportunities to get/in and out. And to do that you are trying the most simpelest thing, an simple EMA crossover strategy backtest over a long period dating all the way back to 1930. And you do it for a couple of EMA lookbacks, just to find out what was more preferable longer or shorter lookbacks. Also you used 3 sectors idexes just to see if there was some stability in the outcome and its not due to 1 lucky sector index. You present 6 performance results of the strategy for the 3 sector indexes for the bull and the bear regime. Your conclusion: esp for the bear regime (peak->trough) the strategy is favourable over buy and hold, eps for ones with shorter lookback. Not so for the bull regime (trough->peak).
My impression for what its worth:
Logical approach?
yes, sure, altough I would add randomized period trails to the result to know what if choosing random regime periods of same length would give a different result. You typically want result of your backtest to be at least 1 to 2 sd better than your random trail average.
Does this research has some value?
Yes I think so, i know investors who would be happy to know about the results. On the other hand the outcome is to some extend not suprising. We all know sharp downturns have sharp rebounds, thats why short lookback EMA entry's are triggered close at the bottom, so buy low will be performing well in stressed markets
But I also had some issues with the article
1) Serious suspicion of future peeking:
-> you can't buy at closing price the same day it has crossed the EMA, because closing price is big part of your weighted price. You should lag the returns, haven't seen any mentioning of this, so I assume it hasn't been done
-> regime definition requires 2 extra months of verification of no break before we called it an official peak or a trough. But in your backtest you use the day of the peak/trough itself in stead of day 2 months later when you actually know its officialy a peak or trough. But I did check only about 5 dates, not all, and excluded the last ones.
2) Lack of observations. It is quite small basis for conclusions on regime effectiveness when we have only ~ 36 regime changes on 3 instruments is about 100 observations. Maybe outside the scope of this article, but when doing serious data science, you really should use the historical index constituence for that. A lot more data and therefore prove how assets react on the different regimes.
3) No analysis of tradingcosts. Esp if you claim EMA(3) to be the optimal, this means you will be going in and out of position. This will heavely effect your performance.
Your own remark in he conclusion this article:
"Data science is the most important discipline for objective market analysis. Mathematics is more important than statistics in market analysis, but data science is more relevant than either."
My conclusion: This article is nice, but not based on data sience, just a couple of backtests and even contains some basic mistakes.