Corporate Bonds

TheTracker

xemeter wrote: »

The Vanguard data table you've chosen cannot be used to show whether a correlation exists between the different categories.

Instead it shows a single data point; the max value within each category.

You would need a minimum of 2 data points from each of two categories to calculate a correlation coefficient.

I accept the quantity of data is important in establishing the significance of calculated correlation. But

a) most people would expect the term "highly correlated" to show less variance, even of peaks, year by year. Look at UK Equities and UK Government Bonds. In 2013 the returns were +21% and -4% respectively. The following year +1% and +15%.

b) diversification across highly correlated asset classes is better than no diversification across such asset classes.

xemeter wrote: »

If you head over to trusnet and compare the indices for developed market equity e.g. North America, Europe, FTSE All Share, FTSE World, you will see that they are highly correlated over periods of 3, 5 and 10 years. This is especially the case over the last 3 years where even the Nikkei 225 index appears to be showing more correlation.

I would lump those four as example in one or max two asset classes. Equities or UK and International Equities.

xemeter

bowlhead99 wrote: »

Forgive my ignorance.

Each of the ten asset categories has fifteen data points. The values for 2000, 2001, 2002, 2003,...2014. Each category (uk equities, em equities, UK index linked etc) has fifteen data points showing the historic performance: the magnitude of the "up" or "down" in every year in series.

So it's a table with 150 figures on it, 15 data points each for the ten categories.

In what sense is that not "a minimum of 2 data points from each"?

Cheers

Try calculating the correlation for any given year, say for example 2010 - its not possible because you only have one data point for each category.

TheTracker wrote: »

I accept the quantity of data is important in establishing the significance of calculated correlation. But

a) most people would expect the term "highly correlated" to show less variance, even of peaks, year by year. Look at UK Equities and UK Government Bonds. In 2013 the returns were +21% and -4% respectively. The following year +1% and +15%.

b) diversification across highly correlated asset classes is better than no diversification across such asset classes.

I would lump those four as example in one or max two asset classes. Equities or UK and International Equities.

a) Assuming that those values were determined at year end, there could have been a point earlier in the year that both values were identical (the sampling point is arbitrary and only taken at year end for convenience). In that case, if a snapshot was taken at that point in time one could then argue that the categories are fully correlated, and that would be highly inaccurate and highly misleading. That's why I'm saying that there is insufficient data within that table to correctly determine correlation.

b) I agree.

bowlhead99

xemeter wrote: »

Try calculating the correlation for any given year, say for example 2010 - its not possible because you only have one data point for each category.

Sorry, you were saying that you needed just two data points in a year and then you could show a correlation, but if you 'only' know the annual performance of all of them for 15 straight years in a row you can't prove anything..? You've lost me.

a) Assuming that those values were determined at year end, there could have been a point earlier in the year that both values were identical (the sampling point is arbitrary and only taken at year end for convenience). In that case, if a snapshot was taken at that point in time one could then argue that the categories are fully correlated, and that would be highly inaccurate and highly misleading. That's why I'm saying that there is insufficient data within that table to correctly determine correlation.

You're right, if you pick and choose your data points arbitrarily with no consistency, you could make the data say all sorts of invalid things about two asset prices that have a random amount of time between the sample date. However, having fifteen annual performance numbers in a row for asset A and the annual performance numbers for Asset B for the same fifteen periods along side, the conclusions can have more meaning.

In doing such a review, it doesn't make a difference to the maths whether the fifteen "periodic return" figures represent minutes, days, weeks, months or years. Some funds are only priced daily, monthly or quarterly for example, and not by the market every minute or every second.

However, the size of the swings between the data points will inherently be larger if you use years in this case rather than days in another, so it's not ideal. And 15 data points is not ideal as it is not a very long series. But, 15 data points is better than two data points, right? Or are you suggesting two per year? In which case you can simply pretend you do have two per year and the analysis is running for 7.5 years.

What am I missing here?