Expert Forecasts of Bond Yields
and Exchange Rates
Jacob Stæhr Mose, Market Operations.
INTRODUCTION AND SUMMARY
In connection with both monetary-policy and investment decisions it is relevant to have a measure of market expectations of the future development in bond yields and exchange rates. Among other things, monetary-policy authorities use such measures to assess the market participants' expectations of future monetary policy. Investors also have an interest in knowing the market expectations, since differences between these and their own expectations are used to formulate trading strategies.
A frequently used method of measuring investor expectations is to compile forecasts of the future value of key financial variables, e.g. the 10-year bond yield in 12 months, based on input from a large number of market participants and experts. The averages of these forecasts are used as a survey-based measure of the average market expectations. From a theoretical point of view, the advantage of using survey-based expectations is that they can be directly interpreted as panel forecasts of e.g. future yield levels. This is in contrast to the implicit expectations derived from the yield curve, which can be complicated to interpret because of e.g. time-varying risk premiums.
This article looks into the accuracy of survey-based forecasts of long-term government-bond yields and exchange rates based on an average of forecasts from, inter alia, international investment banks, cf.
It is concluded that the survey-based forecasts of future long-term yields and exchange rates contain very little information about the future development.
|
COMPILING SURVEY-BASED FORECASTS
|
Box 1
|
|
Consensus Economics Inc. publishes monthly results of a survey where a number of respondents (typically around 30) submit their forecasts of the future value of various macroeconomic and financial variables – including a number of exchange rates and 10-year bond yields in e.g. the
Forecasts by Consensus Economics' expert panel are often cited, and there are also theoretical reasons why an average of many individual forecasts is interesting. Extensive literature2 thus concludes that a combination of forecasts often improves the predictability considerably, even in relation to the individual forecast, that has historically been most precise. |
|
| 1 The respondent groups for the
2 See Timmermann (2005) for an overview of the advantages of combining individual forecasts. |
|
RESULTS
This section compares the accuracy of the bond yield and exchange-rate forecasts from Consensus Economics with the simple forecast that long-term bond yields and exchange rates will remain unchanged over the forecast horizon.
Table 1 shows forecast errors for the consensus forecast and for the "naïve" alternative that no changes will occur. The Table relates to forecasts of the
| FORECAST ERRORS FOR THE
|
Table 1
|
|
| Basis points |
3-month
horizon |
12-month horizon
|
| Mean absolute forecast error |
|
|
| Consensus forecast |
48
|
91
|
| Unchanged yield |
41
|
82
|
| Note: The yield reported is for the 10-year
Source: Consensus Economics and Bloomberg. |
||
Part of the explanation for this remarkable result is that the interest-rate level decreased considerably in the period under review (on average the 10-year yield fell by 26 basis points per year). This falling trend has systematically surprised the panel of experts. The consensus forecast entailed an expected increase in the 10-year yield over a 12-month horizon in 140 months out of 181 (77 per cent). In fact, yields declined in 65 per cent of the months in the period under review. In other words, the experts maintained expectations of rising interest rates during a period characterised by sustained decreases in interest rates.
German 10-year bond yield
Table 2 shows the corresponding results for the German 10-year yield. The picture is the same as for the
| FORECAST ERRORS FOR THE GERMAN 10-YEAR BOND YIELD |
Table 2
|
|
| Basis points |
3-month
horizon |
12-month
horizon |
| Mean absolute forecast error |
|
|
| Consensus forecast |
41
|
90
|
| Unchanged yield |
33
|
73
|
| Note: The yield reported is for the 10-year German benchmark bond. The survey includes 181 monthly observations for forecasts over a 12-month horizon (October 1989 to October 2004), and 190 observations for forecasts over a 3-month horizon (October 1989 to July 2005). Source: Consensus Economics and Bloomberg. |
||
| ACTUAL GERMAN 10-YEAR BOND YIELD AND CONSENSUS FORECASTS OVER A 12-MONTH HORIZON |
Chart 1
|
 |
|
| Note: The yield reported is for the 10-year German benchmark bond. The forecasts relate to October of the year in question. | |
| Kilde: Consensus Economics and Bloomberg. | |
Exchange rates
The comparison of the consensus forecast with the naïve forecast is now repeated for exchange rates. The analysis includes forecasts for the US dollar vis-à-vis the euro[1], the Japanese yen and the pound sterling.
Table 3 shows the percentage deviation in forecast errors between the consensus forecast and the naïve forecast. It is seen that for e.g. EUR/USD the consensus forecast on average gives a 14 per cent higher absolute forecast error over a 12-month horizon than a forecast of an unchanged exchange rate. The consensus forecast has only been slightly more accurate than the naïve forecast in relation to the 12-month GBP/USD forecast. Like the results for the 10-year yield forecasts, the conclusion for exchange rates is independent of the specific choice of criteria for measuring forecast errors.
| Percentage deviations in forecast errors for consensus forecasts and naïve forecasts of exchange rates |
Table 3
|
|
| Per cent |
3-month
horizon |
12-month
horizon |
| Mean absolute forecast error | ||
| Dollars per euro |
27
|
14
|
| Dollars per yen |
14
|
8
|
| Dollars per pound |
6
|
0
|
| Note: Positive values indicate that the mean forecast error is greater for the consensus forecast than for the naïve forecast. The survey includes 181 monthly observations for forecasts over a 12-month horizon (October 1989 to October 2004), and 190 observations for forecasts over a 3-month horizon (October 1989 to July 2005). Source: Consensus Economics and Bloomberg. |
||
The Table also shows that the gap between the accuracy of the consensus forecast and the naïve forecast narrows as the forecast horizon increases. This is consistent with the panel experts' inclusion of insight into fundamental long-term factors such as macroeconomic imbalances in their forecasts. The adjustment of the financial markets to this often becomes evident over a long-term horizon, while the short-term development is more unpredictable.
CONCLUSION
The analysis shows that for 10-year German and
Equivalent results are seen for the three exchange rates over a 3-month horizon, but over a 12-month horizon the panel experts' forecast was in one single case (GBP/USD) more accurate than the naïve forecast.
Overall it can therefore be concluded that the information content of the frequently cited consensus forecasts for US and German yields and three key exchange rates is limited. However, it should also be emphasised that this conclusion does not exclude the possibility that survey-based expert forecasts of other financial or macroeconomic indicators, e.g. economic activity or inflation, may contain useful information about future developments.
The financial markets are often described as efficient. In its strongest form, this concept entails that all available information is reflected in the prices of financial instruments. It is still possible, however, that predictability in monetary policy or systematic risk premiums can lead to a certain degree of predictability of yields and exchange rates. For example, the partial predictability of bond yields is a theoretical implication of the "expectation hypothesis" for the term structure. However, the above analysis does not support the existence of such predictability over horizons of up to 12 months.
LITERATURE
Allan Timmermann (2005), "Forecast Combinations", in Handbook of Economic Forecasting (to be published in 2006).
[1] Before 1 January 1999, DEM/USD is used instead of EUR/USD. The official conversion rate of 1.95583 D-marks per euro has been applied.