Policy interest-rate expectations in Sweden: a forecast evaluation

Abstract

In this article, we evaluate two types of Swedish policy interest-rate expectations: survey expectations and expectations inferred from market pricing. The data are drawn from the most prominent survey of financial-market economists and from Swedish financial markets, and they are carefully matched by date to ensure comparability. Results show that both kinds of expectations suffer from bias and inefficiency, and in terms of forecast precision there is no clear winner. We do find, though, evidence that the forecast accuracy of both kinds of policy-rate expectations has improved since the Riksbank started publishing its own policy-rate forecast, suggesting that this communication strategy has been beneficial from a policy perspective.

Keywords:

• survey data
• monetary policy
• Sveriges Riksbank

JEL Classification:

• E47
• E52

Acknowledgements

We are grateful to Jesper Hansson, Göran Hjelm and seminar participants at the National Institute of Economic Research for valuable comments. The views expressed in this article are solely the responsibility of the authors and should not be interpreted as reflecting the views of the Executive Board of Sveriges Riksbank.

Notes

¹ There is a fairly lively discussion among both academics and policy makers concerning the limits to central bank transparency; see, for example, Morris and Shin (2002), Svensson (2006), Rudebusch (2008) and Ehrmann et al. (2012).

²  Dale et al. (2011) considered inflation forecasts in a highly stylized model but argue that this principle applies more broadly.

³ The Reserve Bank of New Zealand was the true pioneer when it comes to endogenous interest-rate forecasts, publishing endogenous rate forecasts from 1998. Interestingly, though, it publishes its projection of the 90-day bank bill rate, not its own policy rate (the official cash rate). In October 2005, Norway began publishing endogenous projections of the policy rate. Most recently, in January 2012, the Federal Reserve began publishing the end-of-year policy-rate forecasts of individual Federal Open Market Committee (FOMC) members.

⁴ The presence of premia which make forward rates poor predictors of market rates is well documented in the literature; see, for example, Chernenko et al. (2004) and Piazzesi and Swanson (2008).

⁵  Because of the construction of the FRA, it is unsuitable to use the first FRA contract as a measure of expectations at the 1-quarter horizon. We accordingly do not use it in this article.

⁶ Specifically, we observe rates on 3-, 6-, 9- and 12-month overnight indexed swaps which settle on overnight interbank borrowing rates for the following day. These contracts reflect expectations about average tomorrow-next interbank rates over the contract’s life. Converting these rates to forward rates spanning 6 to 9, 9 to 12 and 6 to 12 months, it is then possible to compare the rates to those on FRA contracts that reflect expectations about 3-month interbank borrowing rates. This comparison is made immediately before international monetary market-settlement dates so that the contract horizons match. The difference between the overnight and 3-month forward rates reveals the market’s anticipation of the spread between 3-month and overnight interbank borrowing rates 6 to 12 months ahead, that is, the extra compensation needed for taking on term bank credit risk in the future. We assume that the anticipated 6- to 12-month term credit spread ahead is a good indication of the same spread anticipated at 12 and 24 months ahead. We also subtract a further 10 basis points from the FRA rates (the sample average) which reflects the sample average spread between overnight interbank rates and the Riksbank’s repo rate. Overlapping data on FRAs and overnight indexed swaps are available from September 2002. The data are sourced from Reuters.

⁷ This rule-of-thumb results in an average term premium adjustment similar to that described by Piazzesi and Swanson (2008) for the US.

⁸ Newey–West SEs are used to address the serial correlation in the residuals (which occurs by definition at the 1- and 2-year horizons when using quarterly frequency data).

⁹ In order to investigate this, we conducted some additional regression analysis. At the 1-quarter horizon, there is only one large forecast error associated with the crisis. Allowing for a dummy variable to remove the influence of this, there is still a significant bias for the Prospera data. Detailed results are not reported but are available upon request.

¹⁰ Recall that biased forecasts can be rational, for example if forecasters have asymmetric loss functions; see Elliott et al. (2008) for a discussion.

¹¹ We use the modified test of Harvey et al. (1997) rather than the orignial one suggested by Diebold and Mariano (1995) since our samples are fairly small.

¹² January 2007 is the date of the last Prospera survey conducted before the Riksbank’s change in communication strategy and May 2007 the date of the first survey conducted after this change.

¹³ The one-step-ahead forecast is also the the horizon studied by Dale et al. (2011).

¹⁴ More traditional tests would require a larger number of observations than we have available here.