Emerging Markets - Extending historical returns

[siamond]

In the Simba backtesting spreadsheet, the current data series for emerging markets (EM) is based on:
- MSCI World ex USA index 1985-1987 (Developed Only) - a very crude mapping
- MSCI Emerging Markets index 1988-1994
- Vanguard Emerging Markets Stock Index Fund (VEIEX) 1995+

Quite obviously, this is very imperfect, not going terribly far back in time, plus the 1985-1987 years are too crude to be meaningful.

While reading various studies about emerging markets, the following index history became apparent (here is a reference):
- Early 80s: the first database of EM returns was created by IFC (International Finance Corporation), it was called EMDB (Emerging Markets Data Base). The primary family index, named IFCG (IFC Global), was made available for academics. IFCG numbers became more broadly available between 1985 and 1987.
- 1988: MSCI created its own Emerging Markets index, which has been a primary reference since then.
- 1989: IFC introduced the IFCI (Investable) family index, which measures returns on stocks that are liquid and available to foreign investors (a subset of IFCG, which tracks the most active stocks on a given market, irrespective of availability to foreigners).
- 1999: IFC sold its EM indices to S&P (Standard&Poor), and the indices have been known since then as S&P/IFCG and S&P/IFCI.

Unfortunately, the IFCG numbers for the years predating IFCI and MSCI EM are hard to find.

Fortunately, an interesting graph was included in the Credit Suisse Global Investment Returns Yearbook 2010. Click to see a larger version. Please pay special attention to the description, the authors extended the IFCG EM methodology back to 1976, while staying consistent with the corresponding list and weights of emerging countries. You can also notice that IFCG EM in the early 80s (dark blue line) significantly underperformed MSCI World (developed countries).

[siamond]

For the purpose of the Simba backtesting spreadsheet, using a public index, or something directly derived from an index, works well. And even if such data is imperfect (some non-investable components, most recent research not factored in, etc), it is clearly better than no data, or than a terribly crude mapping. Let's focus on the Credit Suisse 2010 graph then, notably the dark blue line.

It turns out that some online tools are able to extract fairly precise data from such a graph. Having the chart calibrated with its starting point and end point helps a lot, notably for a logarithmic chart. Furthermore, the graph information inside the PDF file can provide additional assistance in calibrating the numbers. A couple of fellow Bogleheads did the heavy lifting for such computer-assisted eyeballing, don't ask me for technical details, but this seems quite effective. We sanity-checked the process by looking at the numbers extracted from the MSCI World line, compared with the actuals, and this worked surprisingly well. While the results will never be perfect using these methods, the error seems reasonable and we're comfortable that the numbers are "close enough" to be educational.

Based on this effort, we can now assemble the following data series:
- 1976-1984: IFCG EM "close enough" annual returns, extended back in time by Credit Suisse
- 1985-1987: IFCG EM "close enough" annual returns, as reported by Credit Suisse
- 1988+: MSCI EM annual returns, then Vanguard

This seems like great progress to me, and worthy of inclusion in the Simba spreadsheet. Any thought, feedback?

[SimpleGift]

It turns out that some online tools are able to extract fairly precise data from such a graph.

Fascinating work, siamond. Can you expand on what these tools are and where they might be found online? Can they be used by the not particularly tech-savvy amateur? Thank you.

[AlohaJoe]

If you're going to extract data from a graph I'd also take a look at the DMS "emerging markets index". I think it was in the 2014 Credit Suisse book. It was 1900-present, so it covers a lot of ground. I seem to recall they made some questionable (but defensible) choices. Emerging markets was any country under $25,000 GDP per capita. I think it was equal weighted.

[dcabler]

It turns out that some online tools are able to extract fairly precise data from such a graph.

Fascinating work, siamond. Can you expand on what these tools are and where they might be found online? Can they be used by the not particularly tech-savvy amateur? Thank you.

Hi Simplegift - There are a number of them out there. One I've used quite a bit is
https://apps.automeris.io/wpd/

And here's a site with a bunch of them compared
http://connectedresearchers.com/graph-d ... your-data/

[siamond]

If you're going to extract data from a graph I'd also take a look at the DMS "emerging markets index". I think it was in the 2014 Credit Suisse book. It was 1900-present, so it covers a lot of ground. I seem to recall they made some questionable (but defensible) choices. Emerging markets was any country under $25,000 GDP per capita. I think it was equal weighted.

Yes, we looked at it too, more precisely the update to the 2014 research that was published in the 2018 summary report from Credit Suisse. In the 2014/18 reports, the DMS authors departed from extending the IFCG index, and created their own data series, using their own methodology. This is what I alluded to in my 2nd post when I spoke of 'more recent research'. We compared the numbers, and sure enough, they are somewhat different, the latest from Credit Suisse being a tad more rosy.

For the goals of the Simba spreadsheet, we felt that the IFCG index numbers were a better fit, if only for consistency with the other data series, which are usually based on a formal index for the early years.

[SimpleGift]

Hi Simplegift - There are a number of them out there. One I've used quite a bit is
https://apps.automeris.io/wpd/

And here's a site with a bunch of them compared
http://connectedresearchers.com/graph-d ... your-data/

Appreciate the links, decabler. I never knew such software existed — and these tools sound like a godsend to amateur financial data sleuths everywhere. I can't wait to try digitizing a few charts! Thank you.

[siamond]

As a side note, for the record, I found S&P/IFCI on Morningstar, Sec ID FOUSA06EUI, starting in 1989:

S&P/IFCI Composite TR USD

Not that it really matters since we have MSCI EM numbers since 1988, but well, this is another data point.

[siamond]

I had gathered bits and pieces, but I finally stumbled upon a paper from S&P that provides full details about the history of IFCG and IFCI (careful readers should check Appendix A). The methodology used for IFCG was somewhat experimental, and refined over time.

https://us.spindices.com/documents/rese ... nload=true

Key extracts:
- The original EMDB Indices were calculated just once a year, used month-end prices and were based on the ten to 20 most active stocks in each of the ten emerging markets. The indices were also equally weighted and available on a "price only" and total return basis. Nine of the ten markets had a history dating back to December 1975; one (Jordan) had a base in January 1978.
- In late 1985, in response to growing interest in emerging markets among the international fund management community, the IFC decided to switch from an equal weighting methodology to one using market capitalization weighting, improve the timeliness of its month-end index calculations by moving from a quarterly lag to a one-month lag, expand the number of stocks covered and increase the number of markets covered from 10 to 17.
- The new indices launched in January 1987 proved very popular with the investment community. Portugal and Turkey were added to the coverage in 1989 and Indonesia was added in 1990.
- The IFC responded [...] by introducing a new set of indices in March 1993. These indices, called the IFC Investable (IFCI) Indices, were designed specifically to be benchmarks for international portfolio managers by taking into account the impact of foreign investment restrictions on the weighting of index constituents.
- When the IFCI and IFCG indices were acquired by Standard & Poor’s (now S&P Indices) in 2000, the index calculation methodology remained largely unchanged. However, the index names were changed to S&P/IFCI and S&P/IFCG, respectively.
- In 2008, Standard & Poor’s – as part of its efforts to streamline its global index offerings – switched from using the chained Paasche method to calculate the S&P/EMDB Indices to the divisor-based methodology used in most S&P indices. This change was implemented on August 1, 2008 for all S&P Frontier Market Indices and on November 1, 2008 for all S&P/IFCI Indices. The S&P/IFCG indices were discontinued effective November 2008. The S&P/IFCI became a subset of the S&P Global BMI (Broad Market Index).

[siamond]

In the Simba backtesting spreadsheet, the current data series for emerging markets (EM) is based on:
- MSCI World ex USA index 1985-1987 (Developed Only) - a very crude mapping
- MSCI Emerging Markets index 1988-1994
- Vanguard Emerging Markets Stock Index Fund (VEIEX) 1995+

Quite obviously, this is very imperfect, not going terribly far back in time, plus the 1985-1987 years are too crude to be meaningful.

Based on this effort, we can now assemble the following data series:
- 1976-1984: IFCG EM "close enough" annual returns, extended back in time by Credit Suisse
- 1985-1987: IFCG EM "close enough" annual returns, as reported by Credit Suisse
- 1988+: MSCI EM annual returns, then Vanguard

Trying to reach a conclusion on this... It seems to me that the Pros outweigh the Cons. Agree? Disagree?

Pros:
- fairly credible approximate data is much better than no data, and we had a big gap in Simba for EM before 1988
- IFCG was *THE* reference index for emerging markets in the early days of such indices (80s)
- Credit Suisse using (and extending) IFCG in their 2010 yearbook is a big endorsement

Cons:
- the design of the IFCG index now seems rather... outdated (equal weights, not all securities were 'investable')
- the IFCG index is no longer maintained by S&P (and the newer IFCI doesn't go far enough back in time)
- extracting numbers from a graph isn't the most reliable approach, even with advanced tools

[siamond]

So...I got three positive answers by private message and e-mails... And zero feedback via posts on this thread... Did I frighten everybody with the arcane historical stuff?

[GAAP]

Sounds to me like you're doing the best you can with the data you can find. The planning engineer in me says you can't hope for much more than that.

[randomizer]

Trying to reach a conclusion on this... It seems to me that the Pros outweigh the Cons. Agree? Disagree?

I think you've summed it up accurately.

[SimpleGift]

Trying to reach a conclusion on this... It seems to me that the Pros outweigh the Cons. Agree? Disagree?

The question seems to be: Is increasing the coverage by an additional 10 years back to 1976 worth the less than perfect data?

Reading a little about the history of the IFC index, it sounds to me like a reasonable representation of what back in the 1970s were called "third world" countries. The index included 10 local stock markets — Argentina, Brazil, Chile, Greece, India, Jordan, Korea, Mexico, Thailand, and Zimbabwe. And though they were apparently equal-weighted, the index tracked total returns (dividends included), plus the representation was fairly broad and at least spread across several continents!

So in my view the pros outweigh the cons.

[siamond]

Thanks for the feedback. Will proceed with the next Simba update. There are a couple of additional ideas floating in the air, stay tuned.

[AlohaJoe]

I would actually vote against it. I think one of the uses of the Simba spreadsheet is to help people see what kind of returns they could have realistically gotten in the past. That's already a bit tricky when we talk about things like "index returns in 1922" when there weren't even really mutual funds yet. But when you start talking about equal weighted third world countries in the 1970s.... That just doesn't sound practical.

What would the expense ratio be? What's the annual turnover?

I guess I'd be less hesitant if we saw the overlap between an equal weighted index and a cap weighted index and decided the difference was tolerable. But from the posted chart they look to be very different.

I can see including the data in an appendix or separate series but I wouldn't splice it into the existing data. Just like it would feel wrong to start using the returns of Powershares Russell 1000 Equal Weight ETF as "total stock market returns".

[siamond]

I would actually vote against it. [..]

I am a little hesitant too, this is why I wanted to have such open discussion. Your points basically elaborate on the list of 'cons' I listed. But you view them as conveying more weight than the other posters so far. I can understand.

This being said, one could limit tests involving EM to 1988+ when using the Simba spreadsheet, and then the fact that the series is spliced to such IFCG numbers pre-1988 would be inconsequential. In any case I would mark such numbers as 'coarse mapping' for sure.

Any other opinion on the matter?

[dcabler]

I'd still vote for keeping it in the main returns section. It's not like this would be the only spliced return series in Simba. Eg, the small cap value series is a splice of quasi-CRSP returns derived from FF data and actual vanguard returns, which itself changed the index it tracks from an S&P index to an MSCI index to a CRSP index. And CRSP's methodology has also changed over the years. Yet I think what's there is just fine as long as you know what its source is as and its limitations.

As noted elsewhere, if you're using Simba as a precision tool to decide on your AA, you're using it incorrectly anyway. I don't actually look at Simba as a tool to tell me what would my returns would have been had I started investing in indices that didn't even exist in, say, 1953. Instead, to me, it allows me to see what big changes might to do my returns trajectory through different economic cycles. We'll never have enough data until t->infinity and I don't plan to live that long. Keep it in, it's good enough for what it's for.

[AlohaJoe]

I would actually vote against it. [..]

I am a little hesitant too, this is why I wanted to have such open discussion. Your points basically elaborate on the list of 'cons' I listed. But you view them as conveying more weight than the other posters so far. I can understand.

This being said, one could limit tests involving EM to 1988+ when using the Simba spreadsheet, and then the fact that the series is spliced to such IFCG numbers pre-1988 would be inconsequential. In any case I would mark such numbers as 'coarse mapping' for sure.

Any other opinion on the matter?

Actually, now that I look at the two charts again they look more similar than I thought. (Maybe I was looking at the MSCI World Index line last time? I dunno.) It would better if the IFCI index were rebased to end-88 IFCG (index of end-88 MSCI World) but just from eye balling it, it appears the IFCI and IFCG have mostly the same big drops and drops. So I guess I've changed my mind and am tentatively in favor of including it. As Tyler said, the simba spreadsheet is good about providing references and explanations so that will hopefully make it clearer how the splicing happened.

[siamond]

It would better if the IFCI index were rebased to end-88 IFCG (index of end-88 MSCI World) but just from eye balling it, it appears the IFCI and IFCG have mostly the same big drops and drops.

Oh yes, the two series are highly correlated, this is clear from the graph.

[Tyler9000]

Actually, now that I look at the two charts again they look more similar than I thought. (Maybe I was looking at the MSCI World Index line last time? I dunno.) It would better if the IFCI index were rebased to end-88 IFCG (index of end-88 MSCI World)...

Using a bit of image editing, here's the same chart with the endpoints aligned. I went ahead and deleted the World line as that's just confusing the issue.

http://oi68.tinypic.com/n6zit.jpg

While the IFCG series diverges a little between 89 and 90, overall I'd call that a good match. I'd vote for including it in Simba, as decent data is better than no data and the spreadsheet is good about providing sources for personal study. Also, good EM data is really hard to come by and this is one of the more reputable older series I've seen to date. Good find, Siamond.

[jc6309]

Earlier I had pm'd Siamon re another matter and had therein mentioned I was in favor of this extension to the EM category. As an afterthought, I should have put a separate message of approval here. (Tyler9000 said it better, "Decent data is better than no data")

[siamond]

Ok, I think we have a consensus here. I'll include this IFCG / Credit Suisse data series for EM 1976-1987 in the next update of the Simba spreadsheet, probably for a mid-year update. I'll mark it as a 'crude mapping' though, if only due to the non-investible aspect of the IFCG index.

Many thanks for the constructive discussion and feedback.