The next step is to evaluate the predictive power of the Business Survey indicator for unrevised and revised quarterly change in GDP. First, it is evaluated in relation to the series included in the indicator. This allows us to examine whether the indicator provides more informative signals about economic developments than the individual survey questions do. The indicator's predictive power is then compared with other survey-based indicators of economic activity. The aim is to test the indicator's ability to paint a current picture of the economy and for this reason this exercise is limited to forecasts for one and two quarters ahead.

The predictive power was tested by adding each indicator and survey question in an autoregressive model of quarterly change in GDP. First, each model has been estimated for the period 2008 Q3 to 2015 Q4 and based on this estimate, the models have generated a forecast for Q1 and Q2 2016. Subsequently, the estimation period has been extended by one quarter and new forecasts have been generated for one and two quarters ahead. The exercise has been repeated until the fourth quarter of 2022^[16] The evaluation period is set at 2016-2022 to give the models enough observations to be estimated (2008-2015) but also enough forecasts to be evaluated. Starting the evaluation period at 2016 avoids an evaluation period centred around the pandemic and allows for a better comparison between revised and unrevised GDP. and consequently the models produce a number of forecasts that can be used to evaluate their average ability to predict GDP growth. This is done by calculating the root mean square error (RMSE) and its ratio to the RMSE from an autoregressive model without an indicator/survey question. A ratio below one indicates that the model generates forecasts that deviate less from the actual GDP growth than does the autoregressive model without indicator/survey question.

The Business Survey indicator has a significantly lower RMSE than most of the individual survey questions for both unrevised and revised GDP data and for both time horizons (see Table 6 in the Appendix). The individual survey questions that perform about as well as the Business Survey indicator are production/sales in the current quarter, new orders in the current quarter, investment plans and access to external financing. However, when the time horizon is extended by one quarter, the forecast errors for the models where these are included increase to a greater extent than the model for the Business Survey indicator.

Despite the fact that the Business Survey indicator is based on a much smaller sample, it performs about as well, and sometimes even better, than the other survey-based economic measures (see Table 3). The forecast exercise also shows that the Business Survey indictor functions as a relatively good predictor for GDP growth independent of GDP revisions, but provides slightly more accurate forecasts for the real-time figures. This is in line with what previous studies on the predictive power of survey-based indicators have shown.^[17] See for example Pichette & Robitaille (2017), Brander et al. (2017) and Chamberlin (2007). The indicator can therefore be assumed to provide valuable signals on GDP growth, regardless of whether unrevised or revised data are being used.

Note: NIER refers to data from the National Institute of Economic Research. SB/SILF refers to data from Swedbank/SILF

Source: The Riksbank.