*3.2. The Main Variables*

As our benchmark measure of financial development, we consider the most used measure in the literature—private credit to GDP (Ibrahim and Alagidede 2017a; Jauch and Watzka 2015; Adeniyi et al. 2015; Nikoloski 2012; Clarke et al. 2006; Levine 2005). Specifically, this variable is defined as "domestic private credit to the real sector by deposit money banks" as a percentage of GDP (World Bank 2012). Private credit does not include credits issued to governments and public enterprises, nor does it include credits issued by central banks. This variable is a common measure of financial dept, which captures the financial sector relative to the economy, and has been documented in the literature as having a strong association with long-term economic growth (Beck et al. 2009). A measure of efficient credit allocation, private credit to GDP, signals the credit worthiness of private institutions, as well as accessibility of the credit market to private individuals (Jauch and Watzka 2015). The mean of the variable for our sample is about 73.5 percent with the range of the variable being from 6 percent to about 300 percent,<sup>6</sup> and the median of the variable is about 65%. For example, countries like Nigeria, Algeria, Pakistan, Mexico, Argentina and Ghana all have less than 22.5% private credit, which is the 10th percentile value of our sample.

The other alternate measure we considered is *private money by deposit money bank and other financial institutions and other financial institutions to GDP*. This is a standard alternate indicator of financial depth that has been used in the finance and growth literature (Beck et al. 2000, 2009). The mean for the variable for our sample is 81.34%, and the median of the variable is about 71%. We consider additional financial development measures as part of robustness analysis, which we discuss in subsequent sections.

#### *3.3. Independent Variable*

Based on Chen (2013), we classify languages that need a dedicated future marking (such as English and French) as a strong FTR language. On the other hand, languages like German and Finnish that do not require dedicated grammar use to mark future events are categorized as weak FTR languages.<sup>7</sup> We construct a dummy, taking the value of 1 for weak FTR languages and 0 for strong FTR languages. We chose this as our independent variable because, as Chen (2013) indicates, agents' intertemporal preferences and decision making are represented via strong and weak FTR languages. Likewise, Mavisakalyan et al. (2018) argue that future tense (for strong FTR languages) can be used to indicate cultural factors, and that usage of such can effect speakers' cognition and behavior (or both).

For our sample of countries, 28.2 percent of observations are assigned a dummy of 1 (indicating weak FTR languages), while the remaining 71.8 are assigned 0 (representing strong FTR languages). Our sample has sufficient regional and continental variation. For example, a number of European countries including Denmark, Belgium, Estonia, Germany, Finland, Iceland and Luxembourg have weak FTR languages. Yet, other European countries

like France, Czech Republic, Latvia, Greece, Italy, Lithuania, Poland and United Kingdom have strong FTR languages.

Following Chen (2013) and Mavisakalyan et al. (2018), the language considered for each country is the major spoken language. Chen (2013) mentions that, for the majority of countries in our sample, there is no intra-country variation in terms of FTR strength. This implies that, in most countries, either one language dominates or a common FTR structure is shared among the languages for multi-lingual countries. As an example of the latter, Chen points to the example of Canada. While the country has significant English and French speaking populations, both are strong FTR languages. Likewise, Mavisakalyan et al. (2018) mentions that since available information on multi-lingual countries is not easily available, checking results with an alternate measure—share of total population speaking a strong FTR language—reduces the sample. As part of robustness analysis, we consider this measure and check our results.
