More than one Y Axis, for correlation study

[Djobydjoba]

Hi Mark,

The Investment Overlay graph purpose is to compare the variation in % of two investments or more. In FM, it's as well the graph to use if we want estimate visually the correlation of two (or more) investments: we can compare with this graph the movements of each investment (peaks and valleys). But, for this purpose (looking visually the correlations), the comparison in % is sub-optimal. The Y scale (% variation) is scaled from the investment with the maximum variation in %, so the other investment(s) can be totally flattened on the graph because in % they don't move so much in comparison.

For example, if we want to compare graphically the correlations between Vix & S&P500, the S&P500 is very flattened, and it's very difficult to see peaks and valleys for the latter:
http://img4.hostingpics.net/pics/372742 ... 3h0602.png


One of the strength of Fund Manager is its ability to manage assets of any type. An investment in FM can be a stock, an option, as well as spread, a relative strength, a sentiment indicator, any other economic data (why not)... So, to compare/study graphically the correlation between investments of different types, each investment should have its own Y scale.

Some examples of graphs with more than one Y scales:
http://img4.hostingpics.net/pics/610052GC10.jpg
http://img4.hostingpics.net/pics/869751 ... 050912.png
http://img4.hostingpics.net/pics/229667 ... 030811.png
http://img4.hostingpics.net/pics/598932 ... y2007b.gif

A graph with 2 Y-scales, to compare two investments, would be great and certainly sufficient.

Another suggestion is the ability to select in the lower part of the TA graph type an other investment (instead of an indicator). So the correlation could be compared too in this context. But this doesn't replace the wish to have a graph similar to the Investment Overlay graph but with absolute values, that is, more than one Y scale.

[Mark]

Hi Djobydjoba,

Understood, and that makes sense. As a work-around, one option you have now is to create 2 graph windows. You can tile the two windows, so they are above/below. Graph one investment in 1 window, and the other in the other window. I realize this isn't as good as your request, but it is at least an option available today.

[Djobydjoba]

Thanks Mark for the work-around tip.

A graph with 2 Y-scales, to compare two investments, would be great and certainly sufficient.

As I re-read, and look at the third example image...
http://img4.hostingpics.net/pics/229667 ... 030811.png

...in fact, be able to compare up to 4 investments in absolute values (4 Y-scales) in the same graph would be indeed very powerful!

Thanks.

[Djobydjoba]

Hi Mark,

I pull up this request. A double Y axis graph is necessary in order to study correlations and divergences between two investments. It is used in many situations to study the markets: between two markets, between an indicator and an index, between a currency pair and a yield spread, etc.

Here are various graphs, picked up here and there, to illustrate some usages:

http://image.noelshack.com/fichiers/201 ... yaxis1.jpg
http://image.noelshack.com/fichiers/201 ... yaxis2.jpg
http://image.noelshack.com/fichiers/201 ... yaxis3.jpg
http://image.noelshack.com/fichiers/201 ... yaxis4.jpg
http://image.noelshack.com/fichiers/201 ... yaxis5.jpg
http://image.noelshack.com/fichiers/201 ... yaxis6.jpg
http://image.noelshack.com/fichiers/201 ... yaxis7.jpg


Thanks.

[Mark]

Hi Djobydjoba,

Thanks for the feedback. I do understand the request, but it seems that the most common use is comparing 2 different items of different plotted values. This seems to be done in a more meaningful way by comparing the relative percentages, like with the "Investment Overlay" or "Portfolio/Investment Overlay" graphs. Another benefit of using relative percentages is you can overlay more than 2 items. Plotting actual values using 2 Y axis can also be more misleading when one of the scales has a much different percentage change than the other. These are just my thoughts, but maybe I'm missing some more useful purposes to plotting actual values on 2 Y axis?

[Djobydjoba]

Hi Mark,

The purpose of using 2 Y axis is only for the study of correlations and divergences between items. It is not suited for performance comparisons, for which the relative percentages are indeed totally suited.

I have to explain better the concept and benefits of studying graphically the correlations and divergences between items.

There are some items that can be very different by nature but are most of the time correlated, positively or negatively. For example, an indicator of confidence can be correlated with a stock index.

When two items ordinarily positively (or negatively) correlated move temporarily in opposite (same) direction, there is a divergence. This can mean something, this can be a leading signal for a reversal to come for the other item, because the correlation has to be restored one day or the other. So, divergences between two items ordinarily correlated are often very interesting to pinpoint.

There are an infinite number of items that can be compared to indexes, currencies, commodities, stocks... in order to study how they are correlated. If we identify that two items are correlated most of the time, this can be a very useful info, in particular when a divergence occurs...

For example, we could compare the indicator "number of Iphone sells" (if available) with the Apple stock price. If we see that the sell of Iphones falls and the Apple stock continues to rise, we could think : hmmm, not good, perhaps there could be a reversal for this stock soon...

There are a lot of well known correlations monitored by the financial operators. If a divergence appears for a "significant period of time", this can be a signal that something is coming.

For example, before each stock market crash there are some sectors and indicators that reverse well before the reversal of the S&P500. For example, the transport sector is well known to be a leading indicator, as the semiconductors sector. And there is another indicator that many operators watch, the Advance-Decline Line. It shows the number of advancing stocks less the number of declining stocks. When the S&P500 is going up and this line is going down at the same time, this can be a signal that the rise of the S&P500 is not healthy. On Google Image, you can see many graphs with the S&P500 and the Advance-Decline Line in a graph with 2 Y-Axis, in order to monitor divergence between the two.

The currency traders / investors often watch closely too the correlation between the currency pair and the yield spread between the 2 year government bond of each currency country, because they are often correlated. And, when there is a divergence, this can be a hint that a movement of the currency pair should come because the correlation has to be restored.

The examples are infinite. More examples (more or less relevant) with the links in the previous post. Graphical study of correlations and divergences can be used for anything. It is possible to try a lot of comparison in order to find interesting correlations. And these studies can really help to better understand the markets and anticipate the next movements.

Now, why the two Y axis and not the relative percentages? Because what we want to study is the correlations and divergences, and not to compare the performances. So we want to have the best possible view of the peaks and valleys (up and down) of each item, even if they are very different by nature. Because, basically, what we only look at is: when this item goes up (or down), is this other item goes up (or down) too? And is it often at the same time, or after a delay (divergence) ?

With relative percentages, the example with the graph I had posted in the first post:
http://img4.hostingpics.net/pics/372742 ... 3h0602.png
The Vix indicator is very correlated negatively to the s&P500, but with the relative percentages we can't study the periods of correlation and divergence between the two because we can't see almost nothing of the peaks and valleys for the S&P500. The maximum amplitude for each item has to be displayed in order to see better peaks and valleys for each item.

Same if we want to study the correlations / divergences of a currency pair with a yield spread, this is not really possible with a percentage scale.
Here's an example (EUR/USD and Yield spread) with percentage scale. Unusable, EUR/USD is a flat line due to the scale!
http://image.noelshack.com/fichiers/201 ... h46-01.png

Depending of the nature of the items, the percentage scale can sometimes be OK to study correlations and divergences, for example if the two items are stock indexes. But most of the time it is not suited, and it is always sub-optimal for this task.

PS: there are advanced softwares where more than 2 Y-axis can be displayed (correlation study with 3, 4 or more items displayed at the same time). Really, 2 Y-Axis (2 items) is OK. More, the graph becomes quickly only a spaghetti plate.

[Mark]

Hi Djobydjoba,

Okay, thanks for the explanation/information. I can see that in the case of just trying to study correlation/divergence using 2 Y axis would provide more resolution. We can consider this for a future update.

[Djobydjoba]

Here are two articles discussing the dual axis (pros & cons, pitfalls, etc.).
It seems there are some interesting thoughts in there.

NB: It seems statistic specialists don't like the dual Y-axis very much.. Anyway, there are undoubtedly frequently used in finance (see the numerous examples posted above) and useful.

https://kieranhealy.org/blog/archives/2 ... wo-y-axes/
http://freerangestats.info/blog/2016/08/18/dualaxes
http://freerangestats.info/blog/2016/08/28/dualaxes2

Kind regards.

[Mark]

Hi Djobydjoba,

Thanks for sharing. Dual Y axis does open up a whole new set of questions...

[Djobydjoba]

Hi Mark,

I have thought again about this request.
(With the links below, if error message "Error 503 Backend fetch failed", please retry, several times if necessary)

Here's another quite interesting article: https://blog.datawrapper.de/dualaxis/
As the other articles posted above, the article is not fond of dual Y Axis neither.
The main reasons of the dislike are: a lot of arbitrary in the chosen scales, the difficulty to read, the misinterpretation.

When looking at the dual-Y Axis graphs we can find on the financial reports / blogs, the scalings seem indeed very arbitrary, and often manually set. For example, on these graphs, the scalings seem adjusted manually in order to start with a tight correlation of the two assets at the beginning of the date range, and to end with a divergence:
- http://image.noelshack.com/fichiers/201 ... yaxis1.jpg
- http://image.noelshack.com/fichiers/201 ... yaxis2.jpg

On these other graphs, the scalings are very creative (arbitrary) again. The scales are stretched or shifted, in order to show something readable or meaningful:
- http://image.noelshack.com/fichiers/201 ... yaxis3.jpg
- http://image.noelshack.com/fichiers/201 ... yaxis6.jpg

I thought that a "basic" independent autoscaling of the 2 axis would be just OK, but it seems that the result could be, in many cases, not very readable nor useful/interesting.


------------------------------------------
As a side note, the free trading plateform TradingView implements the dual Y Axis, so it can be tried.
A screenshot (S&P500 + Vix): http://image.noelshack.com/fichiers/201 ... ngview.png
The steps for Dual Y Axis in this program are:
1) Compare 2) Add Symbol 3) Search and select the symbol to add 4) Settings for the added symbol

(On the screenshot the scaling of the Vix seems not completely maximized, but it's due to a high price not visible with the close price display that has been chosen).

Indeed this is powerful and useful, but probably hard to implement in Fund Manager, which doesn't work that way.
------------------------------------------


Now it seems to me that the dual-Y axis is not so desirable.

There is still though the need to compare two assets with different scale resolutions, with good resolution of peaks and valleys for the 2 assets. The Investment Overlay graph type doesn't work properly in this area.

The comparison of 2 assets one above the other each in its proper single Y scale graph would be OK in fact, and probably better that dual-y axis (simpler, more comprehensible..). This was indeed your suggestion in the first place, but which is not practical in any way currently:

As a work-around, one option you have now is to create 2 graph windows. You can tile the two windows, so they are above/below. Graph one investment in 1 window, and the other in the other window.

The existing 4 displays graphs seems a very good starting point, with already one graph on top of an other. In order to help the comparison of peaks and valleys, the following options could (would have to) be added:

- Replication and synchronization of the cursor in the master graph to the 3 other graphs:
The cursor (track ball or crosshair type) in the "master" graph (that is, the graph that has currently the focus) is replicated to the 3 other graphs. That is, when the cursor is moved in the master graph, 3 others slaved cursors move simultaneously in the other graphs, and replicate the same date. So we could compare precisely the price value in each graph for a given date (with cursor labels displayed in each graph).

- synchronization of the date range
When the date range is changed in the master graph (by keyboard shortcut, mouse selection...), is it changed equally in the 3 other graphs displayed, so the date ranges remain identical in all the 4 graphs.

- Vertically, lock the cursors to the close prices.
This was a suggestion already made here: viewtopic.php?p=17169#p17169
In this context, this option could be required.

- For a better comparison of two graphs it is better to have one graph on top of the other, instead of side by side. But currently the investment stack displays first from left to right, and then top to bottom. So perhaps another option to change this behavior, and to display first from top to bottom, and then left to right?

And perhaps all these options could be activated/desactivated simultaneously in a mode named "Comparison mode" or something...

This is only some ideas / suggestions. I'm sure there are other considerations to take into account. But it seems to me it could be a good way to go to fullfil the comparison need.

Thanks for reading.

PS: this is not directly related, and has already been written in another post, but it would be much neater to me if the the cursor labels would be displayed in the scale legends directly instead of in the graph area.

[Mark]

Hi Djobydjoba,

I'm out of the office this week, so will take a closer look at this when I return.

[Mark]

Hi Djobydjoba,

Thanks for the detailed/thoughtful post. Unfortunately I wasn't able to view any of the images you linked, as I always got 503 errors, even after reloading several times. I get the point though.

Another option would be to add a mode to the Investment Overlay graph such that all plotted lines were not on the same % scale. Each line would still be normalized to 100% at the beginning of the graph date range, but the scale could be set independent for each investment such that each investment reaches the max/min Y axis. There would basically be no Y axis labels, as they would each be at a different scale, but could be useful for a "correlation" mode, as opposed to a "performance comparison" mode which is currently how these operate.

Understood about the cursor labels...

[Djobydjoba]

Hi Mark,

I'm gonna change of image sharing website...

I think your option is described in this article:
http://freerangestats.info/blog/2016/08/18/dualaxes
(in the section "Indexes don’t solve the problem fully")

Unlike what the article writes, yes, this looks like a good solution to compare correlation. And straightforward to use (and I guess implement). It should do the job nicely!

[Djobydjoba]

Hi Mark,

Here's an article about the correlation graphs: https://school.stockcharts.com/doku.php ... n_coeffici.
It shows an interesting and useful correlation graph type:
https://school.stockcharts.com/lib/exe/ ... lewtic.png
https://school.stockcharts.com/lib/exe/ ... oldusd.png

With a dual Y-Axis, and the bottom part that shows the change of the correlation (calculated for a period of xx days) through time.

Analyse graphically correlations is useful in market analysis, trading strategies, and for developing a diversified portfolio. It would be great to have such a feature in FM.

PS: similar / related articles and examples:
https://www.tradingview.com/wiki/Correlation_Coefficient_(CC)
https://www.tradingview.com/ideas/corre ... efficient/
https://www.daytrading.com/correlation
https://www.portfoliovisualizer.com/asset-correlations

[Mark]

Hi Djobydjoba,

Thanks for all the information, and documenting it in one place. That is helpful. We can consider this for a future update.