Difficulties in Modelling the Volume of Illicit Financial Flows

AA's Done the Math So You Don't Have to

In a previous post, AA discussed some of the shortcomings in and misunderstandings about currently accepted estimate for money laundering.

Today we’ll look at issues surrounding modelling these flows.

What can be done to add more precision to the process and to what extent?

Models in General and Their Problems

What follows is a discussion of “direct” modelling.  That is, mathematical models consisting of equations that (a) estimate a current state of affairs or (b) predict one in the future using observable data.

These models are based on assumptions about fundamental processes underlying events. For example, according to most economic theories, lowering interest rates spurs capital investment.  Capital investment spurs increased production, employment, and thus, higher GDP.

Once these relationships are identified the modeler’s job is to quantify the impact of a particular action or development by reducing it to a mathematical relationship (equation).

One example might be a 1% decrease in interest rates will result in a 5% increase in capital investments.  As more and more of these relationships are reduced to equations, a model is constructed.

Observable data are the inputs for the models.

Predictions from the models can be retroactively compared to actual results, providing a feedback loop of sorts.

It sounds very scientific but it is not.

Assumptions about underlying economic processes are often little more than conjectures based on the teaching whether real or imagined of some economic prophet. There’s an interesting book “Economics as Religion”  that describes this process.

It is scarcely better with modeling the value of firms.  Here assumptions are made about growth rates, the risk free rate, risk premia, etc.  If you know this discipline, you know that these models are very sensitive to slight changes in growth and discount rates.

And despite the best efforts of modelers, no model has yet been developed that reliably predicts GDP several years out or the value of a firm.

The point of this is to emphasize that these models and their results are not infallible. 

IFF Models and Their Additional Problems

When we look at modelling IFFs, we see that much of what is available to economic or financial models is not available.

That means that IFF modelling is going to be more difficult and is likely to result in less reliable results.

We don’t have a satisfactory theory or theories that explain the volume of volume of various types of IFFs that can be expressed in mathematical terms.

What drives corruption?

We might say that it is directly related to cupidity and opportunity and inversely related to morality.  Corruption is also dependent on the bribe payer’s cupidity and inversely related to its morality. For both the risk of being caught is a negative factor.

How do we model this?  What is the equation that describes this?

If we could specify these relationships in equations, we don’t have data on the variables in the equations.

How much cupidity is there in the elites in Country A?  If average cupidity in a Country A is X, what is the standard deviation?  It may just be a subset of the elite that engages in corruption.

We have some estimates of data, e.g., ranking of countries for corruption by Transparency International.  But you’ll notice TI call their assessment “Perceptions of Corruption”.

Perceptions seem a slim reed to build a case on.

According to an unscientific poll that AA recently saw, 84% of Twitter users who responded didn’t like the new format.  Should AA have the “perception” that overwhelmingly Twitter users don’t like the new format? NBL! 

We have some single point data from discovered IIF transactions.  But there is no robust set of available statistics.

No actuals we can compare the predictions from of our model to.

In fact, if we had that data, we wouldn’t need the model.

But there are more difficulties.

UNDOC had a working meeting in 2017 to refine its methodology.  Here’s a link to the gateway page with a brief overview and links to more detailed material.

UNCTAD has created a “task force” that held its last meeting this July.

That material outlines two key problems with estimating IIFs with AA’s commentary on each in the “bullet” points immediately below each boldfaced sentence.

There is no single accepted definition of Illicit Financial Flows (IFFs)

1. IIFs range from tax avoidance schemes on legally earned profits to the movement of money arising from illicit activities, e.g., drug and human trafficking, corruption, embezzlement.
2. One of the tasks of the meetings referred to above is to try and develop an accepted definition of IFFs.
3. “Tax minimization” is technically legal. Presumably, transfer pricing transactions between MNC do not belong in IFFs. But how or where does one draw the line between “tax minimization” and “tax avoidance” (clearly illegal)?  Is it a difference of degree or difference of kind?
4. Can we separate out the various subtypes of IFFs from the aggregate total? We want information on IFFs not out of academic curiosity, but to craft policy and further enforcement.  One deals with tax avoidance schemes with a particular set of policies and with drug trafficking or corruption with others.
5. We want to make sure that we don’t double count sub-types in our gross IFFs totals. Are bribes to law enforcement officers and politicians part of the drug trade IFFs? Or do we account for these as part of corruption?
6. Within these high level definitional problems there are some other problems.  When we consider IFFs do we look both at intra-country and inter-country flows?

There is no single accepted method for estimating IFFs.

1. Examples of methods that have been used are (a) analyzing discrepancies in Balance of Payment data and in trade statistics (value differences in the goods traded between two countries), (b) cash to GDP measures (assumption is that cash is the preferred payment method for crime and the informal economy).  Sometimes multifactor models are used in an attempt to compensate for shortcomings in a single factor model.
2. Each of these methods used has drawbacks. We’ll take a look at this topic in a subsequent post.

Alternative Approaches

There are alternatives to direct modelling to estimate IFFs.  But these are likely to be less precise than direct models--if available--would produce.

1. One could for example, focus on a handful of countries which are the largest consumption markets in the world for illicit drugs and estimate what percent of world drug trade these countries represented.
2. Then estimate the annual physical flow of drugs to those countries by making estimates that interdictions are x% of the total shipments to those countries.
3. And then from observable street prices estimate the gross sales proceeds.
4. Then make an assumption about the profit margin to the overseas cartel as well as costs associated with the drugs outside the country of consumption/final sale, making as well  foundational assumptions about the costs the local distributor in the country defrays, e.g. transport, protection, sales and marketing, plus its profit margin.  

As you can appreciate from the chain of assumptions, this sort of alternative model is likely to be less accurate than a direct model.   

The problems with direct modelling and alternative approaches suggests that if new models are developed, we treat their results with healthy skepticism.  They will most likely give directional rather than locational results.

To be clear, that doesn’t mean that we should ignore such models or not try to create such models.

But rather that we not treat their results as incontrovertible “fact”.