A lot of research has been published on the usefulness of equity analysts’ buy and sell recommendations at the firm level. However, far less research has been conducted that addresses the aggregate asset class and country level forecasts of analysts and strategists working for major investment management organisations. Perhaps this can be attributed to the fact that data providers have not maintained quality databases of strategist predictions at the aggregate level. This has changed, because IPE is tracking investment managers’ expectations at the asset class level on a monthly basis, in February 1997. The IPE database consists of 120 asset managers, with more than 7,000 monthly predictions. In the December 2005 issue of IPE we presented the results of our research concerning the equity market predictions. In March 2006 we investigated whether or not bond market strategists add value. In this article we take a closer look at the usefulness of currency predictions.

We will analyse the currency signals of various investment managers in more detail; not as an ‘add-on’ when analysing another asset class, but more as if ‘currency’ were a separate asset class in and of itself.

 

Methodology

We described our research methodology extensively in the aforementioned articles on equities and bonds. We will limit ourselves to presenting a short summary here. In IPE’s Investment Managers Expectation Indicator, managers offer their asset class predictions for the next six to 12 months. In our methodology, every monthly prediction is considered as a separated signal for both six and 12 months, so we calculate the percentage rise (or fall) of the currency for both the six- and 12-month period, and do so on a monthly basis (rolling window) according to the following formula:

 

Dj,k,t * Rk,t

 

with:

j = manager

k = the asset class/region

t = time period

Rk,t = price return (over the six- or 12- month period)

Dj,k,t = [-1, 0, 1] with Dj,k,t = 0 if the manager expects no price change (neutral), Dj,k,t = -1 if the manager expects the price to drop and Dj,k,t = +1 if the manager expects the price to increase.

The overall manager score is then calculated thus:

 

n

å Dj,k,t * Rk,t

t=1

 

The overall six- and 12-month manager scores were then averaged. Next, we rank the n managers from 1 for the best manager to n for the worst manager based on the average overall manager score for all three currencies defined by IPE: $/e, $/¥ and $/£.

Via regression models, we then try to explain these return prediction result rankings through a framework that incorporates the following factors:

q Rank in total assets under management (SIZE) and rank in assets under management in currency overlay (AUM_CUR_OVERLAY) and tactical asset allocation (TAA) overlay (AUM_TAA_OVERLAY) strategies. We also considered the absolute AUM figures (which have of, course, some non-normality problems, but we wanted to see whether they matter by providing us with additional explanatory power that could not be covered by a rank version of the AUM variable). We categorise these six factors (three rank and three absolute level variables) as ‘SIZE-RELATED’ variables.

q We then proceed by looking at the distribution of assets under management within the investment management firm by looking at: % (and rank in %) of equities (%EQ), bonds (%BONDS), money markets/cash (%MMCASH), currency overlay (%CURROVERLAY), TAA overlay (%TAAOVERLAY) and other assets under management, ie hedge funds, real estate, private equity and others (%OTHERAUM). We categorise these as TYPE variables.

q Within the equity component of the investment manager’s overall client portfolios, we then proceed by digging one level deeper by looking at the % of (and rank in %) equities in US, Europe, Japan, Asia and emerging markets respectively (as percentage of firm assets under management: USEQ%, EUREQ%, JAPEQ%, ASIAEQ%, EMEQ%, EQ_OTHER%); we categorise these as EQ_REGIONAL variables.

q In similar fashion, we analyse regional specialisation within the bond market by looking at the % (and rank in %) of bonds in specific bond market segments; public bonds Europe, US and Japan, corporate bonds Europe, US and Japan, other (rest of world) public and corporate bonds and money market products (as percentage of firm assets under management: %PUBEUR, %PUBUS, %PUBROW, %PUBJAP, %CORPEUR, %CORPUS, %CORPJAP, %CORPROW, %FI-OTHER). We categorise these as BOND_REGIONAL variables.

q Obviously, it is also interesting to see if parties that are good in one asset class are also good in another. We do, therefore, incorporate the rank in return prediction quality for the other two main asset classes, equities and bonds, from our previous research (TOTRANK_EQ and TOTRANK_BONDS) within the regression framework for currencies. The idea is to test if there are links between prediction results in these asset classes and the prediction results for currencies. We categorise these as ASSET_LINKAGE variables.

q Next, we look at client types within the group of customers of individual investment management firms. The variables incorporated for this purpose are: % (and rank in %) of different clients served: pension plans (%PENSION), charity and non-profit organisations (%CHARTNONP), family trusts (%FAMTRUST), corporations (%CORP), banks (%BANKS), insurance firms (%INSUR), central banks (%CENTRALB) and government agencies (%GOVERNAG); we categorise these as CLIENT_TYPE variables.

q Last, we distinguish between active and passive mandates/firms (based on the tracking error of mandates) through the variables % (and rank in %) of active assets under management (%ACTIVE), so as to test whether active houses are better in predicting currencies than passive houses.

We also incorporate two dummy variables, one for the country of origin of the asset manager (COUNTRY) and one for the firm’s investment style (STYLE)1. We extract information on all of the aforementioned variables from IPE’s June 2006 supplement, Top 400 Asset Managers. The exchange rates are end-of-month closings from Reuters and MSCI (equally weighted).

 

Empirical findings

In our analysis of the value of equity signals by investment strategists, we found that equity strategists did - on average - add value. The performance of bond strategists was far less convincing. When examining currencies we see that, on average, looking at all three currency pairs for both the six- and 12-month prediction results, scores are slightly more positive. This might indicate that equity and currency markets, known to be more volatile and less efficient than bond markets, are ‘easier’ to predict than bonds when thinking from the perspective of the ‘best’ investment management firms. When looking at the three currency pairs, however, the average total positive result stems mainly from the $/¥ predictions. The results for $/€ and $/£ were, on average, negative. Just as it was the case for equities and bonds, we found quite a bit of dispersion in results among the analyzed managers. The top10 (based on the equally-weighted combined score for the three currency pairs) is listed in table 1.

Table 1 is the third top10 list derived from our extensive research on the value of asset manager predictions (we presented the top10s for equities and bonds in the earlier IPE articles). No asset managers appear in all three lists. Six managers achieved two top-10 classifications: AG Bisset, Aegon, Credit Lyonnais (which no longer exists and is therefore based on older data), Morgan Stanley, Pictet & Cie and Robeco. We have not yet analysed which asset manager would be the best, taking all three asset classes into account using some kind of weighting scheme, or which country has the best asset managers. This will be an interesting area for future research.

Unlike our equity and bond market predictions research, we did not perform any univariate regressions. We had far more data available on individual asset managers (about 65 variables in total), so we decided to opt for a different approach. We started by running (mostly step-wise) multivariate regressions using independent variables from each of the aforementioned factor blocks (or ‘categories’): SIZE-RELATED, TYPE, EQ_REGIONAL, BOND_REGIONAL, ASSET_LINKAGE, CLIENT_TYPE and a step-wise regression for the remaining three variables: %ACTIVE, COUNTRY and STYLE. These seven categories of independent variables were each regressed on the (total) prediction rank (thus taking the prediction results of all three currency pairs into account) and on the individual prediction ranks for the three different currencies as dependent variables. So we started with seven times four (mostly step-wise) regressions to get a feel of which variables matter (ie those that pop up as statistically significant on one or more occasions). We then proceeded with a total step-wise regression including all variables that were important within the category regressions. We will now first summarise the results for the different category regressions in phase one of our empirical test framework. Next, we will analyse the overall multi-category regression framework, after which we will summarise and conclude.

 

Results of the different regressions at the category level

q When looking at the size-related factors, we can conclude that ‘size’ - whether in total assets under management or in currency or TAA overlay programmes - does not matter. We found, however, one peculiarity with respect to the ‘smaller’ currency pairs: $/¥ and $/£. The bigger TAA overlay managers did a relatively poor job. Liquidity problems (in the futures overlay component of their products) for the smaller currencies might be a reason. It could also indicate that these managers are too big to move within these currencies and adjust their opinions accordingly (i.e. they cannot say to their clients that they are bullish/long on a specific currency, while at the same time not buying the accompanying futures to give support to this belief, simply because they are too big as a manager). It could also be that large TAA overlay managers cannot keep the real specialists in-house and that smaller TAA boutiques (often started by people who used to work for the larger ones) are better in smaller currencies.2

q When it comes to TYPE variables (the amount of activity the asset managers have within the various asset categories - in other words, are they specialists or real multi-asset firms?), we found that for both the overall prediction rank and the prediction rank in $/€ it hurts to be an ‘equity specialist’. Being big in the category ‘other assets under management’ (such as hedge funds, real estate and private equity) did help improve the overall rank and the rank for the $/£ currency pair. The $/€ forecasts seem to be better when coming from asset managers with a larger bond than equity component in their set of activities. This corroborates our findings in the March 2006 article on bonds. In that article we documented that bond specialists who also specialise in currencies are actually doing a better job as far as bond predictions are concerned. Since bonds do fluctuate less than equities, the importance of currency fluctuations automatically becomes more important for your overall (international) bond performance. It seems logical that (international) bond houses pay more attention to currency analysis. Apparently this pays off, as when analysing the forecasting quality for the most important currency pair in the world: $/€. For the smaller currency pairs ($/¥ and $/£) this seems less important. For these two pairs it helped to be a currency overlay specialist, with or without strong bond activities.

q With regard to the regional factors (either EQ_REGIONAL or BOND_REGIONAL), there are, firstly, not that many factors popping up as important/significant and, secondly, there is not much consistency among the different currency pairs and total ranks, except perhaps that not having a lot of mandates in US public bonds and/or US corporate bonds might hurt the asset manager’s performance. It seems to us that a substantial presence in these highly competitive markets sharpens the asset manager’s forecasting abilities in the currency area.

q When testing for linkage between ranks in the different asset categories (bonds, equity and currencies), we did not find any significant correlation, although there were more positive linkages between the bond and currency segments than between the equities and the currencies. Table 2 summarises the correlations. Strangely, we did find for the $/¥ negative correlations (albeit for bonds very minor) with both other asset classes, which suggests that being good in either Japanese equities or Japanese bonds is not good for your predictive capabilities for the $/¥.

In our December 2005 research on equity market expectations, we found that being good in Japanese or Asian equities negatively correlates with the predictive power in Western equity markets. This relationship also emerges when looking at the link to currencies.

q With regard to client types, we did not find too many significant correlations. Only in the regression for the overall currency prediction rank did we find a negative correlation for insurers. So, too much of an insurance client base has a derogatory effect on predicting currencies. This is consistent with the US adage that ‘insurers cannot invest’. Another remarkable finding was that having a relatively large client base among charities and non-profit organisations is somehow positively correlated with good currency prediction capabilities. This relationship was not very strong though.

q The three remaining variables (%ACTIVE, COUNTRY and STYLE) were not significant when trying to disentangle the currency forecasts. COUNTRY and STYLE (quantitative or qualitative/fundamental) were not significant. If we have to say anything here, it seems that managers that are 100% active do a slightly worse job than those who also run passive mandates.

 

Overall regressions

After going through the whole exercise of sub-group regressions, we also ran step-wise regressions using a well-diversified list of independent variables that frequently popped up as being significant in the regressions category. We added four variables that did not present themselves as being very significant in the category regressions, but that, according to our qualitative analysis of the output, could still matter in an overall regression framework, namely: TOTRANK_EQ and TOTRET_BONDS and COUNTRY AND STYLE. These variables mattered in our previous research on bonds and equities, and we also wanted to find out if there are any direct links between the asset classes. The results of these multi-category regressions are summarised in table 3. To clarify: for this exercise we worked with 59 asset managers that contributed more than 80 expectations for all three currencies.

The TOT_CUR_Rank regression is relatively straightforward and simple. In the TOT_CUR_Rank regression formula the intercept equals 16.52. This means that if we don’t know anything, we can expect an asset manager to be somewhere around 16-17 out of 59 in our ranking of currency predictions. You then add something to this rank based on the percentage of ‘other assets under management’, the only independent variable found to be significant at the overall level. We use rank variables in such a way that number one represents the highest percentage (or the highest absolute level of assets under management).

When looking at the currency prediction rank at the overall level, the less money an asset manager has in ‘other assets under management’, ie ‘other asset classes or structured finance type of strategies’, the lower his rank (the lower he will be on the list). In other words, with regard to the currency prediction rank at the overall level, it matters more in other asset categories! Taking into account that ‘other asset classes’ incorporates a wide field of relatively new and often complicated asset types (hedge funds, private equity, real estate, commodities, structured finance products) that can be characterised by attracting the brightest minds in the industry, we are not totally surprised.

When looking at the prediction rank for the $/€ we see that %PENSION_Rank (ie, the percentage of pension plans in the client base, with 1 representing the highest-ranked asset management firm (most pension clients), is actually a deduction factor. This means that it hurts to have too many pension plans among your clients. Obviously there is no direct linkage. This relationship represents some kind of proxy, just like the ‘overall assets under management’ did in the overall regressions described above.

In practice, we often see that institutional investors, like pension plans, often demand that their asset managers hedge a substantial part of the ‘traditional’ asset class they manage for the plan. Currency policy is therefore a ‘residual’ of the overall asset management task awarded to the asset manager.

Another negative relationship for the $/€ is the one between rank and ‘size’ as currency overlay manager. It hurts to be a big currency overlay manager (AUM_CURRENCY_OVERLAY_Rank). This might be attributed to the explanation given earlier that overlay managers shouldn’t be too big because market impact might work to their disadvantage. Another explanation could be that, in this specialised asset class, the brightest strategists are more quickly inclined to leave for a smaller boutique (or start one themselves) than they would in mainstream asset classes.

As a third explanatory factor for the $/€ predictive quality, we see that it is actually beneficial to specialise in European equities. This seems to contradict our previous findings for the TYPE variables (see the paragraph on category regressions) that bond managers are doing a better job in predicting currency movements.

This conclusion is in itself not contradicted, but it needs to be slightly mitigated. It can still hurt to be an equity specialist, but European equities are (apparently) an exception.

In the stepwise regression at the multi-category level for the $/¥, we see that a low rank in asset under management in TAA overlays is not beneficial. This contradicts our earlier finding at the category level that it hurts to be a big TAA-overlay manager for the smaller currency pairs $/¥ and $/£. In the former regression with the SIZE variables the AUM_TAA_OVERLAY variable (as AUM in total figures) was significant and the rank variable is significant now.

The former suffers from non-normality problems, so we might have bumped into some statistical artifact here. The relatively low explanatory power of this regression (7% R2) further fuels our suspicion. Then again, when looking at equities and bonds in our earlier contributions we did find that Asia was a tough market to explain with the variables used in our study. This held for bonds and equities, and now our currency analysis confirms the existence of an ‘Oriental mist’.

For the $/£ ranking we posted the largest explanatory power (R2 = 0.40) through a regression with more factors than seen for the other currency pairs. We see that smaller currency overlay managers (ie, specialist boutiques) are doing a better job. On the other hand, having relatively more ‘other assets’ under management helps as well. Managers that also have passive mandates (ie, have a lower rank in the %ACTIVE variable) get higher on the list of good asset managers in $/£. Bond managers also tend to be better, as do bigger TAA overlay managers.

 

Conclusion

When running multi-category regressions, we see that the larger TAA overlay managers are doing a better job for the ‘smaller’ currency pairs $/£ and $/¥, contrary to the initial indications derived from the intra-category regressions, relatively speaking. We do, however, feel that this is not too surprising.

First, when taking other factors into account, we often see regression results changing vis-à-vis outcomes derived from more narrowly defined frameworks.

Second, taking into account the positive relationship with ‘other assets under management’ as well, we feel that the (by broadly oriented definition) TAA manager (a true generalist) is well equipped to handle the complexities of the currency market.

Obviously, currency specialists do a good job too, but it pays to choose a specialist boutique within this asset class.

The linkage between bonds and currencies is interesting as well. If a manager that you choose as currency manager also performs well in bonds, chances are that he will do a good job increase. This holds, albeit to a lesser extent, for European equities as well.

Japan remains, just as we previously showed for equities and bonds, a complicated market. It seems as though we need variables other than the 65 presented in our framework for equities, bonds and currencies to evaporate the ‘Oriental mist’ that surrounds the Asian regression frameworks.

Concentration of insurers (and to a lesser degree pension plans) in the asset manager’s clientele is not a positive indication for predictive capabilities in the currency area. Country and investment management style were not really important; there are more ways to do the job, and parties that do a good job can be located anywhere.

Erik L van Dijk is CIO and partner and Harry J Geels is senior investment manager at Compendeon in Bunnik, Netherlands

 

1 The transformation into ranking variables helped us solve problems related to the non-normal distribution of some variables. This was especially a problem for the skewed assets under management variables. It did not apply to the dummy variables for country (US = 1 and Others = 0) and style (Quantitative/Structured = 1, fundamental/macro = 0, mixed = 0,5).

2 When we talk about TAA overlay managers here, we do not limit ourselves to pure specialists that have no other activity but global tactical asset allocation. We disentangled the activities of all investment management firms in our sample by (sub-) asset class. The TAA component can therefore range from 0 to 100%. Our finding here can be interpreted as follows: a larger TAA component turned out to lead to lower prediction ranks for bigger (in terms of overall SIZE) firms, but not for smaller ones.