MESA bugs on historical data

[Mike77]

Not only real time bugs, but also historical data (time series with more than 2 decimal places) like
- most FX
- spreads and ratios

is not analyzed correctly! (not sure about the dominant cycle)

Not exactly useful when you actually want to apply the methods ...

[jbergeron]

Yeah, those screenshots look horribly wrong; Any news on when these indicators are going to be fixed ESignal?

Time is money

Cheers,
Joshua C. Bergeron
jcb@othernet / #ESignal

[gspiker]

FYI,

I don't have FX data to test with, but historical spreads and ratios work fine here with MESA. I am only running a single instance of each study ... are you running multiple?

G

[TimJ]

I have noticed this same problem (build 582). When I run Mesa with most currencies, I get a square-wave sine indicator. In the GBP (symbol GBP A0-FX) example below, you can see that the sine indicator does not function as designed:



However some currencies plot the sine indicator correctly, as with the Yen below (symbol JPY A0-FX):



Historical and real-time data behave exactly the same in this regard. I have discovered the same thing as Mike - the critical difference is the precision of the currency price.

In general currency prices show 5 digits of precision. For most, like GBP, the price hovers around 1.0. In the first screen shot, the price is close to 1.6379. A price difference of 1/10000 is important on this currency.

Some currencies, especially those crossed with the Yen, have prices typically in the hundreds, for example 116.89. In this case, the trader is interested in price movements of 1/100.

The nature of the bug seems to be related to the precision of the price - the symbols with prices 100 and over, with a precision of 1/100 plot the sine indicator correctly. Symbols with price precision of 1/10000 plot a square wave. I have tested this with numerous currencies and always get this result.

Perhaps this is an error propagation problem in the Mesa algorithm related to precision? Maybe single precision floats being used instead of double?

Anyway, it is easy to replicate, but it sounds like Mesa doesn't have access to FX data to test this. Can we get them access to real-time FOREX please so that they can investigate?

There is a possible work-around for this. If we could display and analyze prices as integers, then Mesa might work correctly. For example, show GBP as 16379 instead of 1.6379. Is there any way to get eSignal to present prices in this fashion?