Does producer confidence predict economic growth in the Netherlands? : a study on the predictive power of producer confidence on economic growth

(1)

1

Does producer confidence predict economic growth in The

Netherlands?

A study on the predictive power of producer confidence on economic growth.

Abstract

This paper investigates whether producer confidence has predictive power on economic growth in The Netherlands. The timeframe from which the data origins is between 1995 and

2016. The methodology used to examine the predictive power of the data has been done with multiple Granger causality tests. The results obtained in this research suggest that producer confidence could play a significant role in the prediction of economic growth in

The Netherlands.

Name: Quirijn Renne Student number: 10444432 Supervisor: dhr. dr. K. Vermeylen Faculty: Economics and Business Specialization: Economics and Finance Date: June 29, 2016

(2)

2

Statement of Originality

This document is written by Student Quirijn Renne who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3 Contents 1 Introduction………...4 2 Literature review………..4-5 3 Data…………..………..5-9 4 Methodology...………..………..9-11 5 Results….………..………12-16 6 Conclusion……….16-17 7 References………..18

(4)

4 1. Introduction

Various confidence indicators are used all over the world in planning the actions of, for example: Investors, manufacturers, retailers, banks and even governments. These businesses can adjust their operations and governments can prepare for changing taxes on the basis of these confidence indicators. Confidence indicators measures the degree of optimism that is felt about various economical topics which is why these indicators are often used to measure the overall shape of an economy in a country. In this research paper the confidence indicators are not used to measure the shape of an economy, but rather it is investigated whether these indicators can be used to predict the economy. Countries, like The Netherlands, have government agencies which conduct surveys that form these confidence indicators. The by far most cited confidence indicator is consumer confidence, but since it has been already researched thoroughly another confidence indicator is investigated in this research paper. Producer confidence, or as some countries call it business confidence, is the indicator that is investigated for its use of predicting the economy. More specific, does producer confidence predict economic growth in The Netherlands?

2. Literature Review

In the past multiple research papers have investigated the same or similar topics as will be done in this research. In this literature review these papers will be examined on the basis of how the research is conducted, what is different compared to this research and what were their results.

The research of Taylor and McNabb (2007) examines whether both consumer and producer confidence indicators can predict movements in GDP. This analysis focuses on four European countries, UK, France, Italy and The Netherlands for the period from 1983 to 1998. In this research the time frame is different so there might be different results. The results they obtained show that both confidence indicators have good predictive power in identifying turning points in the business cycle, especially in the UK and The Netherlands the indicators reduce the forecasting error associated with the prediction significantly. These results encourage the research which will be done in this paper, in which the business confidence indicator might be a predictor of growth of GDP and not just the turning points in the business cycle. “Taken as a whole, our ﬁndings suggest that conﬁdence indicators might usefully supplement macroeconomic models and forecasts of economic activity, hence potentially aiding policy makers.” (Taylor and McNabb, 2007, p. 205). Taylor and McNabb used the cross-correlations among variables to determine whether the confidence indicators can help with the prediction of movements in GDP.

A second research has been conducted by Carnazza and Parigi (2003). In their research they investigated the ability of business confidence indicators to predict the short-term evolution of GDP in Italy with the Granger causality test. The business confidence indicator is based on a survey and is published monthly. Since the evolution in GDP in Italy is published on a quarterly basis the confidence indicator has to be altered in order to perform a valid Granger causality test. The business confidence indicator in The Netherlands, just as in Italy, is also published on a monthly basis so the same alterations as Carnazza and Parigi are a possible solution. In their research two different approaches were found in order to transform the

(5)

5

monthly dataset to a quarterly dataset. The first approach is a more common one since it is generally adopted by the European Commission and other institutions. The quarterly data is formed by the arithmetic mean of the three months in the quarter. The second approach Carnazza and Parigi used in their research is basing the quarterly data on the three-term moving average centered on the last month of the quarter. This is basically the same as the first approach but instead of the mean of the three months in the quarter, they take the mean of the last two months of the quarter and the first month of the next quarter. This approach has been adopted by the ISEA (Institute for Studies and Economic Analysis, kind of the Italian Central Bureau for Statistics). In this research both approaches will be tested although according to them the different approaches do not seem to affect the results significantly, the ISEA approach appears to be slightly better (Carnazza and Parigi, 2003, p. 590 - 595). The reason both approaches are tested in this research is because of the different time frame and because their research is based on data from Italy instead of The Netherlands, their findings do not necessarily have to be the same in this research paper. Beside these two approaches, four more approaches are investigated which I have not found in previous research but might help with the prediction of economic growth. Carnazza and Parigi found in their research that both indicators actually do have predictive power on the evolution of GDP in Italy.

A similar research has been conducted by Santero and Westerlund. They did research on multiple confidence indicators and on multiple macro-economic variables for more than 20 countries, under which The Netherlands, which makes it a broad research paper. In their research they found that in the Netherlands business confidence Granger causes GDP, while GDP does not Granger cause business confidence. This would mean that business confidence is relevant in the prediction of GDP. This research has been done in 1996 and to my knowledge is still a working paper, so the results in this paper could easily deviate from their results.

3. Data

In order to calculate the prediction power of producer confidence on economic growth in The Netherlands, data has to be acquired about both. The Centraal Bureau voor de Statistiek or Central bureau for statistics (CBS) conducts a questionnaire every month in order to measure producer confidence. It’s used as a confidence indicator for the short term development of the industrial production in The Netherlands. This confidence indicator is based on the collection of sample data from 1800 different business establishments, with 5 or more employees, every month. A questionnaire is sent to these businesses with questions about their expected activity in the coming three months, their opinion about their order position and their opinion about their stock position. The data of this research is published monthly and is openly available on their website. The CBS distinguishes producer confidence for a number of different industries and also for the total of all of these industries. In this research paper only the total of all of the industries will be used. After the collection of data the CBS removes multiple biases, for example seasonal influences. Producer confidence has been developed by the European Commission as a so called Industrial Confidence Indicator. The CBS in its turn publishes since the beginning of 1997 the producer confidence data according to the same approach as the European Commission suggests. The indicator is comparable

(6)

6

with other confidence indicators in The Netherlands, like the far more popular Consumer Confidence, which both come from the same dataset from the CBS about the economic cycle. If a company fails to react to the survey, its confidence will be estimated by the CBS accordingly.

Producer confidence is the simple average of responses to questions about these business establishment’s opinion on their current stock position, their current order position and their expected activity in the coming three months as compared to their normal positions. These questions can be answered positively, negatively or neutral, the percentage of negative replies is subtracted from the percentage of positive replies to get a percentage which ranges from -100% (all negative replies) to +100% (all positive replies). This is done with the replies on all three questions and the average is the producer confidence for that month.

As can be seen in the graph 1, the total of all of the industries producer confidence are shown for every month in the period between February 1985 and March 2016.

The main explanatory data, the total producer confidence, has 374 monthly observations in the period between February 1985 and March 2016. The following table shows the modalities of the whole dataset of producer confidence.

-25,0% -20,0% -15,0% -10,0% -5,0% 0,0% 5,0% 10,0% 15,0%

Graph 1. Producer Confidence, The Netherlands

Feb 1985 - Mar 2016

(7)

7 Table 1. Producer Confidence dataset information.

Producer confidence

Timeframe Feb 1985 - Mar 2016 Number of observations 374

Publishing Frequency Monthly

Mean 0,28%

Median 0,60%

Standard Deviation 5,06%

Minimum value -23,50%

Maximum value 9,40%

The second dataset needed is economic growth in The Netherlands. This datais also openly available on the website of the CBS. In this research economic growth is defined as the percentage growth of the real gross domestic product. In The Netherlands it is published by the CBS every quarter which means the percentage growth equals the growth in a quarter of a year in comparison to the quarter before. Below, in Graph 2, the growth of real GDP is shown for the period between the second quarter of 1996 and the fourth quarter of 2015.

Comparing the graphs of both datasets it seems clear that they look similar and probably a high correlation exists between the two variables. The similarities between the graphs does not necessarily mean that producer confidence has predictive value on economic growth. The growth in gross domestic product in The Netherlands is available quarterly for the period between the second quarter of 1996 and the last quarter of 2015, this would mean there are 79 observations in total. -4,00% -3,00% -2,00% -1,00% 0,00% 1,00% 2,00% Q 2 1996 Q 1 1997 Q 4 1997 Q 3 1998 Q 2 1999 Q 1 2000 Q 4 2000 Q 3 2001 Q 2 2002 Q 1 2003 Q 4 2003 Q 3 2004 Q 2 2005 Q 1 2 00 6 Q 4 2006 Q 3 2007 Q 2 2008 Q 1 2009 Q 4 2 00 9 Q 3 2010 Q 2 2011 Q 1 2012 Q 4 2012 Q 3 2013 Q 2 2014 Q 1 2015 Q 4 2015

Graph 2. Percentage Change of Real Gross

Domestic Product,

(8)

8 Table 2. Economic Growth dataset information.

Real GDP growth

Timeframe Q2 1996 - Q4 2015

Number of observations 79 Publishing Frequency Quarterly

Mean 0,47%

Median 0,50%

Minimum value -3,20%

Maximum value 1,70%

A problem arises when comparing the two data sets: producer confidence is published on a monthly basis and economic growth is published quarterly. An interesting thing about producer confidence is that one third of producer confidence comes from the questions about expected activity the coming three months, the rest arises from their current position in stock and orders. Given this fact a possible approach to the problem might be to take the producer confidence of the last month of the previous quarter as well as the first month of the current quarter each time in order to most accurately predict economic growth. This would mean that the producer confidence data set is altered in a way that only the producer confidence of the months March, June, September and December are shown in the first alternative and the data of the months January, April, July and October in the second alternative. In order to try and predict the coming quarter as accurate as possible both of these alternatives will be tested. Since producer confidence is one third of the expected activity of the coming three months, the same approach will be done with the producer confidence value excluding the opinions about current positions of stock and orders. The expected activity dataset is also published by the CBS and is obviously different from the whole producer confidence dataset, the dataset has the following modalities and graph. The graph also shows kind of a similar form as the other two graphs which would again indicate a high correlation so this approach might also help with the prediction of economic growth.

Table 3. Expected Activity dataset information. Expected Activity

Time Frame Feb 1985 - Mar 2016 Number of observations 374

Publishing Frequency Monthly

Mean 0,29%

Median 0,90%

Minumum value -34,00%

(9)

9

Another approach is mentioned in the research of Carnazza and Parigi (2003, p. 590). In this research they adopted a more common approach which has generally been adopted by the European Commission and other institutions. Their approach is based on the arithmetic mean of the three months in the quarter. Their second approach is using the three-term moving average centered on the last month of the quarter, this approach has been adopted by the ISEA. Which means the arithmetic mean of the last two months of the quarter and the first month of the next quarter. So, in total there six approaches to alter the producer confidence dataset to solve the monthly to quarterly problem in this research paper. The hypotheses will be formulated as follows:

Hypothesis 0. Producer confidence does not have predictive power of economic growth in The Netherlands.

Hypothesis 1. Producer confidence does have predictive power of economic growth in The Netherlands.

4. Methodology

To accurately test if a time series dataset has prediction value of another time series dataset is to conduct a Granger Causality test. The Granger Causality is a combination between an ordinary least squares test and an F test. The Granger causality test tests whether lagged values of variable x could improve the ability to predict variable y after controlling for lagged values of y. If this is true the test concludes that variable x Granger causes variable y. In here x would be one of the producer confidence indicators and y would be economic growth. The first step is to conduct an ordinary least squares test for which the y value is in this case economic growth and the x values are the lagged values of producer confidence. In this research paper the number of lagged values will be either four or eight because we use

-35,0% -25,0% -15,0% -5,0% 5,0% 15,0% 25,0%

Graph 3. Expected Activity, The Netherlands

Feb 1985 - Mar 2016

(10)

10

quarterly data and the lagged values will be a full year or two combined, it’s economically not logical to use a number of lagged values which do not add to a full year, also previous research mentioned in the literature review used four lagged values. The unrestricted regression has the following form:

In here ΔGDP is the economic growth of a quarter, PC is the producer confidence value for the quarter it represents, t is the time value, n is the number of lagged values and ɛ is the error term. The second step is to conduct another regression but now only with the lagged values of economic growth as the restricted version, the formula will look as follows:

∆𝐺𝐷𝑃𝑡= 𝛼 + 𝛽1∆𝐺𝐷𝑃𝑡−1+ ⋯ + 𝛽𝑛∆𝐺𝐷𝑃𝑡−𝑛+ 𝜀

The third step is to conduct an F test, for this we need the following information about the datasets and the regressions: The number of observations in each data set (n), the dependent variable plus the number of independent variables in the unrestricted regression (k), the number of independent variables that have been removed to get to the restricted regression (m), and the sum of squares residual for both the restricted and the unrestricted regressions (SSEr, SSEu). With these values an F test can be conducted by the following formula:

𝐹 𝑉𝑎𝑙𝑢𝑒 =(

(𝑆𝑆𝐸𝑟 − 𝑆𝑆𝐸𝑢) 𝑚 ⁄ ) (𝑆𝑆𝐸𝑢 (𝑛 − 𝑘)⁄ )

Using the F value and the right degrees of freedom, k - 1 and n – k, a probability value (P value) should come out of the F table. Preferably this value should be below the significance level of five percent, which has also been used in previous research. If the P value is below five percent it is said to be statistical significant, which in this example would mean that producer confidence has a significant effect on economic growth and can help with the prediction of economic growth. Now if this is the case, you can also say that the producer confidence indicator Granger causes economic growth in this example.

Now a second granger test is conducted for each producer confidence indicator in order to find out whether variable y also Granger-causes variable x, or economic growth would also Granger cause producer confidence. The reason to also do this second Granger causality test is to see whether the variables used are intertwined or if the predictive value only goes one way. This can be done by doing the opposite test as has been done above, the first regression will look like this:

Note: The only thing that changed is the y value, the regressors are the same, this is the unrestricted version of the F test again. For the restricted version the ΔGDP values should be removed, this will look like this:

∆𝐺𝐷𝑃𝑡 = 𝛼 + 𝛽1∆𝐺𝐷𝑃𝑡−1+ ⋯ + 𝛽𝑛∆𝐺𝐷𝑃𝑡−𝑛+ 𝛾1𝑃𝐶𝑡−1+ ⋯ + 𝛾𝑛𝑃𝐶𝑡−𝑛+ 𝜀

𝑃𝐶𝑡= 𝛼 + 𝛽1∆𝐺𝐷𝑃𝑡−1+ ⋯ + 𝛽𝑛∆𝐺𝐷𝑃𝑡−𝑛+ 𝛾1𝑃𝐶𝑡−1+ ⋯ + 𝛾𝑛𝑃𝐶𝑡−𝑛+ 𝜀

(11)

11

This will be the restricted version of the F test. Just as above conduct an F test with the same value for n, k and m since the same datasets and lags are used as in the first regression. Only the sum of squares residual for both the restricted and the unrestricted regressions (SSEr, SSEu) is different since it is another regression. This would naturally mean a different F value and, after using the same degrees of freedom in the F table, also another P value. Again the same five percent significance level is used in this F test. If the P value is under the significance level it is said to be statistical significant, which in this example would mean that economic growth has a significant effect on producer confidence. If this happens economic growth could help predict producer confidence. If it occurs that both the first and second P values are statistical significant both values can help with the prediction of the other.

Now the tests described above will be done in multiple variations, for example with four lagged values as has been done in previous research and also with eight lagged values. To solve the problem for monthly to quarterly data in the producer confidence data set, six possible solutions have been mentioned before, the first two by Carnazza and Parigi, the arithmetic mean of the three months in the quarter, and the three-term moving average centered on the last month of the quarter. The other four variations come from the fact that a third of the producer confidence indicator comes from the questions about expected activity about the coming three months, the rest comes from their current position in stock and orders. It might be useful to try and predict economic growth on the basis of the expected activity, and thus use the producer confidence of the last month of the previous quarter. So the producer confidence dataset is altered in a way that only the producer confidence of the months March, June, September and December are shown in these tests. Another variation is the use of the first month of the current quarter instead. This would mean that the data of producer confidence of the following months is used: January, April, July and October. These two variations are the same approaches as the last two variations, but then excluding the opinions on orders and stock. This would make six different producer confidence indicators To complete the analysis the tests are done on again three different alterations of these six producer confidence indicators. The first is of course the plain vanilla dataset, the second is using the change compared to the previous quarter indicator instead of the indicator itself. The third variation is a little more complicated, it arises from the fact that the confidence indicators in general are released well in advance of the data on economic growth. The data on economic growth can take up to 45 days after the quarter has ended to publish, and this number has to be revised two times in the same year so it is not definite. This fact makes it very interesting to predict the current economic growth which hasn’t been published or revised yet with the help of the already published confidence indicator. This would mean that there are in total eighteen different producer confidence indicators for which the Granger tests will be done with four and eight lagging values.

(12)

12 5. Results

The results of the Granger Causality tests will be presented in a table for each of the six different producer confidence approaches. In the first column the Indicators are categorized in two sets of three indicators each, the set with four lags and the one with eight lags. In each set are three different alterations of the producer confidence approach used in that table, the first is the unaltered dataset, the second is the same dataset but with a lag in real GDP to try and show the predictive value of the current quarter producer confidence on the current quarter growth in real GDP. And finally the third is instead of the ordinary producer confidence the tests has been done with the change of producer confidence compared to the previous quarter. In the second and third column the P values of the Granger causality tests are shown, these P values represent the probability that one variable does not Granger cause the other. In this research paper a significance level of 5% is used which means that the P value should be below 5% in order to reject the standard null hypothesis of the Granger causality test: Indicator does not Granger cause economic growth. In the last two columns are regression analysis of the very first ordinary least squares test done which is part of the Granger Causality test, the R2_{and the standard error of the regression are shown here. These}

two numbers can show how good of a predictor the producer confidence indicator is compared to the other indicators.

To start off with the first set of tests, in here producer confidence is the arithmetic mean of the three months in the quarter. This approach has generally been adopted by the European Commission according to Carnazza and Parigi.

Table 4. Results Granger Causality tests of producer confidence as the arithmetic average of the three months.

Immediately it stands out that every different indicator of producer confidence actually Granger causes the growth of real GDP, the P values of these tests are all below 0,1% which would mean that it is below the significance level of five percent. The null hypothesis in the second column is rejected for all indicators. To start off with the tests with four lagged values in the regression. The first producer confidence indicator which is the plain vanilla one, in which it is just the arithmetic mean of the three months, has the lowest P value (3,12%) in the third column. This value would mean that growth of real GDP would also Granger causes this

Indicators

Null hypothesis: indicator does not Granger-cause GDP

Null hypothesis: GDP does not

Granger-cause indicator

Standard error of the regression Number of lags: 4

Producer Confidence 0,05% 3,12% 0,484 0,00547

Producer Confidence t+1 0,00% 11,73% 0,613 0,00477

Producer Confidence Change 0,09% 79,68% 0,475 0,00552

Number of lags: 8

(13)

13

producer confidence indicator with the significance level of 5%. The other two four lagged indicators have no significant P value in the second Granger causality test, which would mean that growth of real GDP does not Granger cause these producer confidence indicators. The P values of the Granger causality test with eight lagged values are all significant. In here the producer confidence indicator Granger causes economic growth and vice versa. The highest R2_{and lowest standard error of the regression for four and eight lagged values both come}

from the producer confidence t+1 indicator, this would mean that its predictive power is the highest. Which is no surprise since this indicator uses the current quarter producer confidence to predict the current economic growth. For clarification purposes, the t+1 indicator with four lags regression looks like the following, please note the t instead of the t-1 in the first producer confidence regressor.

The next table shows the results of the producer confidence indicators based on the three-term moving average centered on the last month of the quarter. This is the approach which has been adopted by the Institute for Studies and Economic Analysis (ISEA), the Italian central bureau for statistics according to Carnazza and Parigi.

Table 5. Results Granger Causality tests of producer confidence as the three-term moving average centered on the last month of the quarter.

Again, just as the results in the first table, all of the producer confidence indicators seem to Granger cause economic growth here. The P values are even lower than in the first table, they are all below 0,01%. The second Granger causality tests in the third column of the table are for the four lagged tests all not significant. This means that economic growth does not Granger cause these producer confidence indicators. The P values in the third column with the eight lagged indicators are all around the significance level of 5%, the plain vanilla indicator and the t+1 indicator are just below 5% and the producer confidence change indicator is just over 5%, thus only the first two are Granger caused by economic growth. Just as in the first table, not surprisingly, the highest R2_{values and lowest standard error of the}

regression came from the t+1 indicator. The R2_{and standard error of the regression are even} Indicators

Number of lags: 8

Granger Causility tests Regression analysis

∆𝐺𝐷𝑃𝑡 = 𝛼 + 𝛽1∆𝐺𝐷𝑃𝑡−1+ 𝛽 ∆𝐺𝐷𝑃𝑡− + 𝛽3∆𝐺𝐷𝑃𝑡−3+ 𝛽4∆𝐺𝐷𝑃𝑡−4+ 𝛾1𝑃𝐶𝑡+ 𝛾 𝑃𝐶𝑡−1 + 𝛾3𝑃𝐶𝑡− + 𝛾4𝑃𝐶𝑡−3+ 𝜀

(14)

14

higher and lower compared to the ones in the first table. It seems that just as Carnazza and Parigi concluded in their research that the ISEA confidence indicator slightly outperforms the confidence indicator adopted by the EC in the predictive value of economic growth. This too is not surprising since the ISEA indicator is made up of the average of the last two months of a quarter and the first month of the next quarter. So the first month producer confidence of the quarter it tries to predict is used for the prediction of that same quarter.

The following two tables, table 6 and table 7, present the results obtained from the producer confidence indicators based on the producer confidence of the last month of the previous quarter and the first month of the current quarter.

Table 6. Results Granger Causality tests of producer confidence as the last month of the previous quarter.

Table 7. Results Granger Causality tests of producer confidence as the first month of the current quarter.

To start off with plain vanilla indicator, ‘producer confidence’ in the table, there is one outlier in table 7 with four lags for which it is not significant that the indicator Granger causes economic growth. Its P value equals 15,83% which is well over the significance level of 5% and thus the null hypothesis is not rejected. What is an interesting outcome is that this indicator does get Granger caused by economic growth, with a P value of 0,04%. The producer confidence t+1 indicator again Granger causes economic growth but this time economic growth also Granger causes the indicator, which was not always the case before. The producer confidence change indicator show in the second column high P values, except for the one in table 6 with eight lags. A possible explanation for this is that the change could be

Indicators

Number of lags: 8

Indicators

Number of lags: 8

(15)

15

very high since the producer confidence is based on a single month and the change indicator would be the difference between that month and the month a quarter away. Since a lot can change in a quarter this dataset is probably very volatile which translates to the low R2_value

and high standard error of the regression value. The R2_{and the standard error values}

compared to the first two tables are unfavorable.

The last two tables, table 8 and table 9, present the results obtained from the producer confidence indicators based on the expected activity of the last month of the previous quarter and the first month of the current quarter.

Table 8. Results Granger Causality tests of producer confidence as the expected activity of the last month of the previous quarter.

Table 9. Results Granger Causality tests of producer confidence as the expected activity of the first month of the current quarter.

The results of tables 8 and 9 look a lot like the results of the tables 6 and 7. This is obvious since the expected activity is one third of producer confidence as a whole. The same reasoning as for tables 6 and 7 could be interpreted on these results as well. It is interesting to see that the R2_{and standard error of the regression are both not as favorable in the results of tables}

6 and 7 as expected. These values differ only a very little in each indicator, sometimes even in favor of just the expected activity. The question arises from this is that the opinion of these business establishments on their current stock and orders is relevant for the prediction of economic growth. These results would suggest that these opinions are probably insignificant

Indicators

Number of lags: 8

Indicators

Number of lags: 8

(16)

16

since it differs so little. A second interesting finding is the fact that the producer confidence indicators based on the last month of the previous quarter seem to outperform the producer confidence indicators based on the first month of the current quarter. This is concluded from the R2_{and standard error values and is true for the producer confidence indicators and the}

expected activity indicators. It is interesting because it is expected that the producer confidence of a month in the quarter it tries to predict would outperform the producer confidence of the last month of the previous quarter.

A very clear pattern emerges that has also been found in the tables of results with respect to the t+1 indicator. The producer confidence t+1 indicators seems to always Granger cause economic growth and it has again the highest R2_{value and the lowest standard error of}

regression for both the four and eight lags. It is expected of this indicator since it predicts economic growth of a quarter using the producer confidence of that quarter, but it is still interesting to actually see it in all six different approaches to producer confidence on a quarterly basis.

6. Conclusion

As noted in the literature review, not a lot of research has been done in the past on producer confidence, consumer confidence is by far more popular. The previous literature that has been done have shown that producer confidence indicators could help in the prediction of economic growth. Unfortunately there is no research available in more recent years. The datasets used in this research with the timeframe of 1995 to 2016 is in my opinion a big enough sample to be able to do the tests done in this research. Also the fact that the data used in this research is published by the CBS in The Netherlands and that the producer confidence is made on the basis of rules by the European Commission since 1997 show that the data is trustworthy and of a certain quality. The Granger causality test is widely used and reliable, but it has its limitations, for example when both X and Y are driven by the same third variable with different lags, the null hypothesis might still be failed to reject.

As for the results obtained in this research, according to the R2_{values and the standard error}

of the regression the best method to alter the monthly producer confidence dataset to a quarterly dataset is by the three-term moving average centered on the last month of the quarter. This method has been adopted by the ISEA and it shows with significance that producer confidence does Granger cause economic growth and would thus be able to help with the prediction of economic growth. Comparing these results to the producer confidence based on the average of the three months in the quarter, as adopted by the EC, it seems that the ISEA indicator performs slightly better when looking at the R2_{and the standard error}

again. This is no surprise since the ISEA indicator uses the first month producer confidence of the quarter it tries to predict and the EC indicator does not. As can be seen in the tables 4 and 5, all of the indicators Granger cause economic growth which will mean the original null hypothesis: “Producer confidence does not have predictive power of economic growth in The Netherlands” is rejected. The ISEA end the EC indicators have shown their ability to help the

(17)

17

prediction of economic growth with four and eight lagged values, with the plain vanilla dataset, the t+1 dataset and even with the change in producer confidence dataset.

Now for the results obtained in this research about producer confidence and expected activity which are based on a single month. The R2_{values and standard errors would suggest that}

there is minimal difference between the use of producer confidence or expected activity when basing it on a single month. This leads to an interesting possibility, is the opinion on current stock and orders of the business establishments even relevant for the prediction of economic growth? But comparing the R2_{and the standard errors to the ISEA and EC indicators}

it suggests that they would have significant less predictive power, also because not every indicator Granger causes economic growth. Another interesting finding is the fact that producer confidence based on the last month of the previous quarter performs better than producer confidence based on the first month of the current quarter, for both producer confidence and expected activity. This is also based on the R2_{value and the standard error.}

This is even more interesting since the findings of the ISEA indicator outperforms the EC indicator because the ISEA indicator uses the first month of the current quarter too. This contradiction is not tested so it might just be an interesting coincidence.

(18)

18 References

Batchelor, R., Dua, P. (1998). Improving Macro-economic forecasts: the role of consumer confidence. International Journal of Forecasting, 14, 71 - 81

Carnazza, P., Parigi, G. (2003). Tentative Business Conﬁdence Indicators for the Italian Economy. Journal of Forecasting, 22, 587 - 602

Fisher, K. L., Statman, M. (2003). Consumer Confidence and Stock Returns. The Journal of

Portfolio Management, 30(1), 115 - 127

Granger, C. (1969). Investigating causal relations by econometric models and cross-spectral methods. Econometrica, 37(3), 424 - 438

Green, R. (1997). Follow the Leader How Changes in Residential and Non-residential Investment Predict Changes in GDP. Real Estate Economics, 25(2), 253 - 270 Mourougane, A., Roma, M. (2003). Can conﬁdence indicators be useful to predict short term

real GDP growth? Applied Economics Letters, 10, 519 - 522

Santero, T., Westerlund, N. (1996). Confidence Indicators and their relationship to changes in economic activity. Working Paper

Sum, V. (2014) Effects of Business and Consumer Confidence on Stock Market Returns: Cross-Sectional Evidence. Economics, Management, and Financial Markets, 9(1), 21 - 25 Taylor, K., McNabb, R. (2007) Business Cycles and the Role of Confidence: Evidence for