Why do I see "Trends" in Randomly Generated Data?

[Indrionas]

Quote from MBAGearhead:

Attached is simple line chart with random data. Just repeatedly hit F9 key in Excel and you'll get a new chart each time. You'll see so many trends in random data it will make your head spin!

No Java needed.

Yes, this is good. I only wanted to show a little more sophisticated way of generating random charts: OHLC data, not just a line chart of closing prices.

 

[Indrionas]

Quote from NZDSPeCIALISt:

Thanks for posting this. Could you generate and post more charts like this (like the one without the gaps)?
It looks so "real".

Why do I see tradeable "trends", "candlestick formations", "trendlines", "chart patterns",
etc. etc. in randomly (within the parameters you have chosen) generated data??

Yes, I can post a few more later if you wish

I think that when one thinks about "tradeable trends", he just fools himself by applying trend-following concept to the parts of the data and ignoring the rest of data where the trend-following concept fails miserably, especially when you include trading costs.

 

[SoCalTrader619]

Do you think you'd be able to modify this Excel worksheet to show OHLC?

Quote from Indrionas:

Yes, this is good. I only wanted to show a little more sophisticated way of generating random charts: OHLC data, not just a line chart of closing prices.

 

[Jerry030]

Quote from Indrionas:

In case you want to feed the OHLC data of the above chart into some hardcore analysis tools, here's the data in the attachment.

Try fibonacci numbers, moving averages, oscillators, ACD, support and resistance lines, fitting ARCH model, try something more sophisticated, like neural networks and genetic algorithms.

I'm curious. What value is there in the analysis of pure random data?

What relationship does that have to a market where humans have money at risk?

I'd say none and time spent on such efforts is a fools errand... unless one believes the market is random, and even if that is the case, why not look at real market data?

I mean in medicine a lab test would involve the blood of the patient who is sick not the blood of random healthy people or "synthetic" blood.

If you been analyzing this kind of thing it would explain why your efforts have failed.

 

[Indrionas]

Quote from Jerry030:

I'm curious. What value is there in the analysis of pure random data?

What relationship does that have to a market where humans have money at risk?

I'd say none and time spent on such efforts is a fools errand... unless one believes the market is random, and even if that is the case, why not look at real market data?

I mean in medicine a lab test would involve the blood of the patient who is sick not the blood of random healthy people or "synthetic" blood.

If you been analyzing this kind of thing it would explain why your efforts have failed.

You just completely misunderstood my post. It was just a sarcasm.


The next chart shows what happens when a tiny bias is introduced. It was generated using the same process (without gaps) and a 1% of upside bias, which means probability of up tick is 50.5% and down tick is 49.5%. In this case you see a real trend. Although the fluctuations "around it" are random, and "buying on dip" is as good as buying at any other random point.

 

[Indrionas]

Here's OHLC data for the above chart.

 

[NZDSPeCIALISt]

filter, filter, filter...

Quote from Indrionas:

Yes, I can post a few more later if you wish

I think that when one thinks about "tradeable trends", he just fools himself by applying trend-following concept to the parts of the data and ignoring the rest of data where the trend-following concept fails miserably, especially when you include trading costs.

keep those charts coming!

 

[Indrionas]

Quote from NZDSPeCIALISt:

filter, filter, filter...




keep those charts coming!

Here, only three charts, because creating random data charts is quite boring and dull work (I prefer video games)

You can see there are lots of "tradeable trends", lol

 

[fundjunkie]

Quote from Indrionas:

It is possible. In that case you would need to find out what is the "ticks per day" distribution in your market, and if there is a bias in up tick versus down tick. Then, with this information you can generate thousands of possible markets with the same characteristics as your target market and see where it ends up, for example, 95% of the time, aka Monte Carlo @95% confidence level. The only question: is such analysis applicable in practice?

Well, I'm thinking here about generating more data for backtest purposes.

 

[Indrionas]

Quote from fundjunkie:

Well, I'm thinking here about generating more data for backtest purposes.

Then you are assuming stationarity and the system will only work on the same market conditions that you are simulating. But market data is not stationary and distributions change over time.