EMU, Diversification, and LoCorr2
By Herbert Blank

Diversifying internationally in the face of increasing correlations among countries and even continents is a problem now and may well become more so in the future. Blank tackled a particularly thorny example, a client needing the right kind of large cap diversification using European stocks for a large cap US portfolio. Blank came up with a solution that points the way toward broader research, and maybe similar solutions beyond both large cap and Europe.

Aplan sponsor of a major, multibillion fund approached me earlier this year - let's call him George. George was viewing the portent of European Monetary Union (EMU) with alarm. The more data George received from his consultants, the less separation he could find between recent price patterns exhibited by European stocks and patterns exhibited by US stocks. Even within Europe, the individual countries' stock markets were becoming increasingly correlated with each other. Surely, the accelerating trend of European markets performing more similarly would only be exacerbated by the advent of EMU.

The pension consulting industry has long counseled plans to earmark at least 10% of their plans overall assets for international equity. Studies have shown that the combination of exposure to potentially higher returns and low correlations with U.S. equity returns improve the overall risk/reward profile of the fund. Recent trends, however, have severely challenged the wisdom of this strategy. Yet George wanted diversification, and for a variety of reasons, he wanted it from Europe.

THE DECLINE OF COUNTRY FACTORS

The real problem with increasing correlations among the equity markets of countries within Europe goes back to the basic rationale for investing in markets outside the United States. The combination of desirable rates of return combined with low correlations to US markets makes foreign markets attractive to US portfolio planners. In fact, the more different asset classes that can be combined with distinctly different up-and-down patterns, the more diversifying into many different countries' markets could potentially improve the risk/reward ratio of the overall retirement fund. Therefore, to the extent that the stock markets of European countries behave in similar manners, the benefits for an equities portfolio of diversification among countries in Europe will be reduced.

The decrease in relative importance of country factors changes European stocks' price behavior. From the end of 1988 to the end of 1997, the average correlation of the countries who have committed to the European Monetary Union ("EMU Countries") during the prior 156 weeks rose dramatically from 0.49 to 0.61. Pronounced, distinguishable country factors appear to be giving way in relative importance to an EMU-precipitated "Euro-factor."

DETAILED RISK ATTRIBUTION ANALYSES

A more detailed analysis of European correlation can be developed through factors used by BARRA in risk attribution. BARRA has been a leading resource of the pension community for more than 20 years; it is a leading provider of risk management and performance attribution software. BARRA decomposes historical performance data to quantify how sensitive each security has been to different capital market factors. This analysis concentrated on three BARRA factors: country of issuer; industry of security; and market capitalization. To identify and compare how significant a determinant each factor had been over time, the time interval from January 1, 1986 through December 31, 1997 was divided into four separate periods. Then, for each factor, a regression analysis was performed against each security's price change. From this series of regressions, T-statistics were determined for each factor during each period.

Using the same time periods and intervals, the significance of the BARRA country factors for the EMU countries has declined from a T-statistic of 6.0 to a T-statistic of less than 3.0. Conversely, during the same period, industry factors as a differ-entiator of stock returns increased dramatically. The T-statistics of BARRAindustry factors during the same time period for the average EMU country rose from 2.4 to nearly 4.

Even more pertinent to the problem at hand, "lcap"- the large cap factor within BARRA - has enjoyed a substantial rise in prominence. In the initial period (1986-88), the average T-statistic of lcap for EMU countries as a determinant of stock returns was 0.32 - not at all significant. During the next three-year period (1989-91), lcap was even less significant for this group of countries' equity markets, with a T-statistic of 0.13. However, the past six years tell quite a different story indeed. The lcap factor during the 1992-1994 period proved very significant in return differentiation, with a T-statistic of 4.6. In 1995-1997, the significance of lcap grew even more dramatically, as evidenced by an average T-statistic for EMU countries exceeding 9.8. Finally, using a three-year interval ending June 30, 1998, the T-statistic for lcap was 11.1, breaking 10 for the first time.

Moreover, the rise in significance of lcap has been paralleled in the five developed European country stock markets (Denmark, Norway, Sweden, Switzerland, and UK) who have not yet committed to joining the European Monetary Union. The average T-statistic for lcap among the "non-EMU five" was 0.38 in the initial period and 8.0 in the most recent period. Examining the behavior of this group of five markets with respect to the other two BARRA factors we examined yielded complementary results, albeit less dramatic. The average T-statistic of country factors for the non-EMU five has fallen from 5.8 in the initial period to 3.5 in the most recent one. For industry factors, the analogous T-statistics, were 2.9 and 3.8 respectively.

PROBLEMS OF LARGE CAP DOMINANCE

Returning to the case of George, the sudden rise in lcap as a determinant of European stock returns exacerbates his diversification dilemma. If the country of domicile for a European equity is a less significant determinant of return than its market capitalization, then investing in large cap European stocks does little to diversify the risk profile of his overall portfolio. This is especially true because the bulk of George's US-equity portfolio consists of investments in large cap US stocks, including a large S&P500 Index fund.

Monthly correlations between the European component of the Dow Jones Global Indexes (DJGI-Europe), which was used for this work, and the S&P 500 Index from January 1, 1992 through June 30, 1998 measure 0.731, indicating a significantly strong tendency for stock prices of both indexes to move in tandem.

Having validated George's concerns, the question remains: Is it possible to construct a viable European stock portfolio which provides a significantly different risk profile than the S&P500? The most obvious empirical answer, a portfolio consisting exclusively of European small cap stocks, is not a practical solution for George. Significant liquidity issues and ownership restrictions render unfeasible a significant commitment by him of funds to this market segment. The data make it clear, also, that concentrating investment in non-EMU countries will not contribute to significantly better risk diversification.

THE SIMULATION

One potential approach was to control portfolio risk through the employment of risk optimization software. The risk model selected for the analysis was the APTModel devised by Dr. John Blin for his company, Advanced Portfolio Technologies, Inc. The idea was to create a suitable European portfolio that is relatively uncorrelated with the European component of the Dow Jones Indexes (DJGI-Europe) and, by extension, the S&P 500 Index. The next step was to construct a backtest with procedures simulating actual portfolio behavior as it would have occurred during the simulation period as precisely as possible. Accordingly, the following simulation procedures were employed:

1. Start With Worldscope Database

2. Eliminate Issues With Market Cap < $250 MM (US)

3. Eliminate Passive Foreign Investment Companies and Restricted Shares

4. Leave In "Dead" Companies As Eligible For Selection/Inclusion to avoid "look-ahead" bias

5. Using APTFactor Analysis, Create Risk Bogeys

6. Create Bogey As It Existed at Beginning of Each of the 26 Quarters From January 1, 1992 Through June 30, 1998

7. Create Reasonable Turnover Dampening Rules

8. Formulate Realistic Transaction Cost Assumptions

9. First Run: No Tilt File

10. Run APTOptimizer to Get Each Quarter's Results

11. Second Run: Try a Tilt File To Enhance Performance

12. Rerun APTOptimizer to Get Each Quarter's Results

RESULTS

From a pure risk diversification perspective, the new portfolio series (LoCorr1) achieved its goal. Not only did each portfolio have a significantly different APT risk profile from DJGI-Europe and the S&P 500, but the low correlations of monthly returns with both benchmark indexes were dramatic. The correlation between LoCorr1 and the DJGI Europe during the 26 month period was 0.003. The risk diversification offered between LoCorr1 and the S&P500 is even more dramatic with a slightly negative correlation between the two of -0.036. So, the simulation indicates strongly that the goal of using only European stocks to create a portfolio that will provide risk diversification with respect to the S&P 500 and DJGI-Europe is achievable.

Unfortunately, there is no investment rationale for LoCorr1 beyond risk diversification. This is because of its performance. For the time period measured, $1.00 invested in LoCorr1 would have grown to just $1.40 in contrast with the $2.28 that was achieved by DJGI-Europe and the even more robust $3.17 posted by the S&P500. So, although LoCorr1 demonstrated that the risk diversification goal is achievable, George would have an exceedingly difficult time selling performance of this nature to his investment board.

Naturally George inquired what would happen if a tilt was utilized in an attempt to increase portfolio performance. Again, the Worldscope database was utilized in order get the necessary fundamental data on the eligible securities. The experimental tilt adopted was a 50-50 combination of low price/book ratio and high 12-month price momentum (both with a 3-month lag). This is a combination strategy that has demonstrated historical effectiveness with U.S. securities according to other studies, including James O'Shaughnessey's, "What Works On Wall Street."

 Everything was done as before. The only difference is that the optimizer was now tilted towards the value-momentum variable. The portfolios produced each month were still optimized to produce an APT risk profile neutral to the DJGI-Europe's risk exposures, but this time with equal emphasis on investment considerations. Terming the new portfolio LoCorr2, we ran the identical tests for the identical time periods, This time the results were more encouraging.

In an unprecedented boom period for U.S. stocks with respect to the rest of the globe, LoCorr2 - composed completely of European stocks - still did not fare as well as the S&P500. However, LoCorr2 would have grown each dollar invested into $2.61 in the 26-quarter period measured as opposed to the meager $1.40 managed by LoCorr1. This is also somewhat better than the $2.28 attained by DJGI-Europe during this period.

Notably, the excess performance of LoCorr2 over LoCorr1 did not come at the expense of diversification. The monthly correlation of LoCorr2 was 0.001 with DJGI-Europe and -0.040 with the S&P 500. Therefore, the success of LoCorr2 demonstrates that it is, at least theoretically, possible to use European stocks to deliver risk diversification and satisfactory performance at the same time.

More research will surely prove useful here. One potential problem is an implicit look-ahead bias. The tilt variable was not data-mined in the classic sense; various permutations of in-sample data were not examined in order to find the best conceivable tilt for the period. Rather, the general construct of the tilt variable was based upon knowledge that this variable had some success in the U.S. stock market. Other considerations involve implementation shortfall concerns and time-period-specific dependency.

Despite these disclaimers and limitations, it seems clear that investors concerned whether sufficient diversification will be provided by European investment in the Post-EMU world should investigate the usage of risk optimization software.