[MUSIC] In this section I will explain you how we proceed with normalization, the waiting and the aggregation of the variables to get to the final index. The first step includes the normalization of all variables. We do so because we want to compare and combine variables, which initially are measured in very different units. Normalization also accounts for outliers in the data. We normalize a variable by ranking all observations according to their values using the percentiles of the distribution. The resulting scale ranges from 1, the minimum value, to 100, the maximum value. Note that we normalize each variable over the panel and not over the cross-section. This approach, which is called panel normalization, has the advantage that it ensures comparability over time. A drawback of the method, however, is that the values for the past change when, for example, an additional year is added. This is sometimes difficult to communicate to users of the index. Here we see the distribution of the trade to GDP ratio. While most of the countries have a trade to GDP ratio around 100%, a few countries have very high values. Singapore, for example, has a trade to GDP ratio that exceeds 400%. The distribution is thus highly right skewed. After normalizing the variable according to the percentiles of the distribution, we obtain a uniform distribution. Each observation is assigned to a value from 1 to 100. I will show you the panel normalization in a small example in which we are looking at three countries in four years. Take again the trades to GDP ratio for three countries, Romania, Germany, and Singapore, for the years 2004 to 2007. Thus, we have a sample of 12 observations and we'll rank them according to the decentiles, where the smallest value is one and the highest value is ten. In our complete sample, we can normalize according to percentiles, because we have more than 100 observations. Here we see the trade to GDP ratio for the different countries and years. While Romania and Germany have similar ratios, the ratio for Singapore is much higher. When ranking these variables according to the decentiles, we obtain the following normalized values. Singapore, in 2006, with the highest ratio of the whole sample, receives the maximum value of 10. Germany, in 2004 and 2005, with the lowest ratio, receives the minimum value of 1. The other observations are ranked in between according to the distribution. After normalizing the variables, the next step is to determine the weights with which each variable enters the index. We apply principle component analysis, PCA, to do that. By using PCA, more weights are assigned to variables that have more variation over years or countries and bear more information about the overall variation in the data. PCA is applied in three subsequent steps, one step for each level of the index. Using this procedure, we obtain the weight of each variable, each of the six subgroups, and each of the three dimensions of the index. A caveat of this method is that weights are fixed over time. This might pose a problem if the relative importance of variables change over time. Variables that used to represent globalization well in the 80s and 90s might not be that important anymore today. For example, letters sent internationally have gradually been replaced by emails. Ideally, we would have a procedure that allows for dynamic weights. The weights are used to aggregate the values of each variable to the final index. I will show you the aggregation, using again the example of the trades to GDP ratio. The trade to GDP ratio enters the subcategory, Actual Flows, with a weight of 22%. Actual Flows, together with Restrictions, sum up to the Economic Globalization, with a weight of 50% each. Economic, Social and Political Globalization together form the overall KOF Globalization Index. The weight of Economic Globalization is 36%. Given all these weights, we can calculate each aggregation level of the KOF Globalization Index based on the individual variables. If data on one variable is missing for a given country, we adjust the weights of the remaining variables within the same subcategory and assign higher weights to the remaining variables. If 40% and more of the variables are missing, the subcategory is not published for that country. Note that this does not imply that we cannot compute the overall globalization index for this country. This is because we calculate every aggregation level based on the individual weighted variables. After normalization, waiting and aggregation, we have calculated the KOF Globalization Index. Savina will now present you the results from the 2016 vintage of our index. [MUSIC]