More on Open Border policies around the world, Part One

As I have described in previous posts, I have been looking at outliers for open border policy around the world.

Today I did a lot of work on it, improved my data, and now have a database of 208 borders in the world, with 124 different countries represented.

Let's dig into the data.

To refresh, my variables are:

• Corruption Perceptions Index
• Press Freedom Index
• Ease of Doing Business Index
• Homicide Rate
• Democracy Index
• GDP
• Population
• GDP per Capita
• The name of the open border treaty the country is in.

From this I have derived the  differences and averages for every country pair for each of these variables, and made a viability score, with the CPI, PFI, DBI, and homicide rates for every country averaged appropriately, with a higher viability score meaning a country which is more likely to have an open border.

Regressional Analysis

Ordinary Least Squares

Most importantly, what determines an open border? Well, after cleaning the data, dropping null values, and building a Ordinary Least Squares Regression, I was able to predict which countries had an open border with an R-squared accuracy of 0.589. The significant variables at 95% accuracy were the difference in GDP per capita (closer GDP per capita means you are more likely to have an open border) at a t score of 3.159. Ease of Doing Business difference at -3.697, an average Corruption Perceptions Index at a t score of 5.064, and finally an average homicide rate at a t score of 2.236.

In other words, be bordering countries with similar GDP per capitas, similar Doing Business scores, minimal corruption, and few murders. Given this, my model predicts you will have an open border.

Random Forest Regressor

A Random Forest Regressor was able to predict with 79% accuracy. The most important feature was having a low Corruption Perceptions Score at 0.5, followed  by having few homicides at 0.11. I'm seeing a trend here.

Random Forest Classifier

My Random Forest Classifier found having a free press, then low corruption, followed by few murders were the best predictors for having open borders.

Conclusion from Analysis

I am finding that safety and freedom are the two internal factors which are predictive of having open borders with your neighbors.

Corruption

I then sorted out all of the country pairs which have open borders leaving only those which have closed borders, and then sorted them by their corruption score.

The country pair with the highest average CPI (which is good) and a closed border is the United States and Canada (79.5), followed by Singapore and Malaysia (67.5). there are currently no plans to open the borders in these two country pairs yet. The next was Saudi Arabia and Qatar (61.5) which used to have open borders. All other closed borders in the world have a corruption perceptions index below 60.

Press Freedom

Sorting by Press Freedom, we again find the United States and Canada have the highest average press freedom (19.505) out of all country pairs with closed borders in the world, followed by:

• South Africa and Namibia (20.315)
• Argentina and Uruguay (20.805)
• Ghana and Burkina Faso (20.87)
• Costa Rica and Nicaragua (22.210)
• Costa Rica and Panama (22.285)
• Botswana and Namibia (22.765)
• Botswana and South Africa (22.840)

There are a good number of countries with a lot of press freedom which could open borders according to this metric.

Ease of Doing Business

When it comes to economics, the Ease of Doing Business Index doesn't seem to be a major factor when determining having an open border with your neighbors, but for interest, here are the results:

• Singapore and Malaysia (82.920)
• United States and Canada (81.005)
• Georgia and Azerbaijan (80.960)
• Malaysia and Thailand (79.525)
• Armenia and Georgia (79.325)
• Turkey and Georgia (78.805)

The Caucuses have made incredible progress on economic liberalization, which is part of why their economies have grown so rapidly over the last decade, particularly Georgia. The US and Canada is again one of the top candidates according to this metric. Singapore and Malaysia are Asian Tigers, so it is not surprising they appear on this list, as well as Thailand.

Homicide Rate

This is where we are really going to hit a rut in terms of potential visa liberalization.

• Croatia and Slovenia (0.760)
• Bhutan and China (0.875)
• Qatar and Saudi Arabia (0.940)
• Bulgaria and Greece (0.945)
• Ghana and Burkina Faso (1.025)
• Vietnam and China (1.070)
• Tajikistan and China (1.115)
• Bulgaria and Romania (1.195)
• Singapore and Malaysia (1.215)
• Croatia and Serbia (1.215)

This is the only list the US and Canada does not appear on because the United States homicide rate is so high. A visa free zone among ASEAN countries might be on the horizon, as this shows. Almost every country pair which has a low homicide rate and other variables which are potentially important already have an open border already. Croatia is in the process of joining the Schengen area, as is Bulgaria. There is no statistically significant impact of homicide on whether a country has an open border or not.

Corruption Perceptions Index

When I construct a decision tree classifier on my data, trying to predict the status of whether any two bordering countries have an open or closed border, I am able to predict whether a country has an open or closed border with 81.25% accuracy. This is really good, and I find that the Corruption Perceptions Index predicts 49.7% of the weight.

The less corrupt your country is, the more likely you will have an open border. The country pair which you see around 80? That's the United States and Canada at 79.5, followed by:

1. Malaysia and Singapore, 67.5
2. Saudi Arabia and Qatar, 61.5
3. Botswana and Namibia, 58
4. Israel and Jordan, 57

Average Score

I then calculated the average score by the equation Average CPI * Average Doing Business Index / Average Press Freedom Index * Average Homicide.
This gives us a relative score which tells us which countries have the highest likelihood of making open borders, we can also see which variables the countries can work on to make it a reality. The countries which do not have open borders which this model predicts are top candidates are:

• Croatia and Slovenia (Croatia is likely going to join Schengen soon)
• Ghana and Burkina Faso
• Bulgaria and Greece
• Canada and the United States
• Singapore and Malaysia
• Croatia and Serbia (Serbia is applying for EU membership)
• Romania and Serbia (Serbia is applying for EU membership and Romania is going to join Schengen soon)
• Bulgaria and Romania (both will likely join Schengen soon)
• Qatar and Saudi Arabia (currently closed to dispute on Qatari-Iranian relations)
• Croatia - Hungary (Croatia is likely going to join Schengen soon)
• Romania - Hungary  (Romania is likely going to join Schengen soon)
• Bulgaria - Romania (Bulgaria is likely going to join Schengen soon) 

Looks like my model is performing fairly well.  I get a 0.579 Pearson R squared coefficient between this model and the reality.

More Interesting Graphs

This makes it pretty clear that more democratic countries tend to be less corrupt, and tend to have an open border with countries close to them with similar freedom metrics, as well as wealth. We see that all of these variables are correlated to each other tightly, which leads us to a fundamental question in political economy, how do wealth and freedom effect each other?

This question is extremely important, because it has enormous implications on all types of domestic policy. For countries at the low end of the spectrum which are poor, authoritarian, and corrupt, will improving their economy lead the way to political reform or entrench the powers that be? Dr. christian Houle answered this question in his paper Inequality, Economic Development, and Democratization which he shared on his personal blog. The reality seems to be that inequality, economic development, and democratization all interplay with one another, and where one stands on these three scales implies different paths to the others.

This is a good overview for one blog post, and I intend on continuing this exploration later.

My backworks is available on my Gitlab repo