Proposed methodology for estimating global money laundering

Known incidents of money laundering involving large amounts of money generated from crime, are of tremendous public interest and are consequently given wide publicity. A wide range of national and international agencies have attempted to quantify organised crime and components of money laundering in their particular sphere of interest, and their assessments are frequently made available in public statements. A comparatively simple crime-economic model, constructed from readily available international databases, closely predicts a range of such expert assessments, and appears to offer a framework for determining and monitoring the size of money laundering flows around the world. Further research is required to complete the model, but the nature of that research is made clear, and it appears that existing data sources are likely to be adequate. Initial output from the model suggests a global money laundering total of $2.85 trillion per year, heavily concentrated in Europe and North America. This web page gives a general overview of the methodology and some interim results. Presentations made in June 1999, to the KriminalExpo in Budapest (with special emphasis on Eastern Europe) and to the "Cyberlaundering" conference in Trento (including global estimates of money laundering via the Internet), are also available.

Comments on this model are welcome and should be sent to born1820@ozemail.com.au.

Background

In early 1998, the retiring chairman of the O.E.C.D.'s Financial Action Task Force (FATF) Working Group on Statistics and Methods, Mr Stanley Morris, stated that "the need to estimate the size of money laundering and quantify its constituent parts has been a concern of the FATF since its initial report."

His report identified at least four areas of legitimate demand for quantitative measures of money laundering:

• Understanding the magnitude of the crime, so that law enforcement authorities, national legislators, and international organizations can reach agreement on the place of counter-money laundering programs within national and international enforcement and regulatory agendas.
• Understanding the effectiveness of counter-money laundering efforts, by providing a baseline and a scale for measurement and enabling evaluation of particular programs or approaches.
• Understanding the macro-economic effects of money laundering, particularly the adverse effects of money laundering on financial institutions and economies. E.g. changes in demand for money; exchange and interest rate volatility; heightened risks to asset quality for financial institutions; adverse effects on tax collection and, ultimately, on fiscal policy projections; contamination effects on particular transactions or sectors and behavioral expectations of market actors; and country-specific distributional effects or asset price bubbles.
• Understanding money laundering, since even the rigorous examination of the components of measurement should produce a deepened understanding of the relationships among, and the differences between, various parts of the phenomena that are grouped together when we speak of money laundering.

He concluded however that "There is not at present any economic deus ex machina that will allow the accurate measurement of money laundering world-wide, or even within most large nations. The basis for such estimations simply does not exist". Almost two years after FATF's quest for quantification began, the Working Group and its economists – as if trying to prove the old theory about laying economists end-to-end - have yet to reach a conclusion on a methodology.

Introduction

This paper begs to differ from Morris's gloomy assessment and describes a logical crime-economic model, resembling an inter-regional input-output economic model, that uses a range of publicly available crime statistics to estimate the amount of money generated by crime in each country around the world, and then uses various socio-economic indices to estimate the proportions of these funds that will be laundered, and to which countries these funds will be attracted for laundering. By aggregating these estimates, an assessment can be made of the likely extent of global money laundering, and comparisons can be made of each country's contribution to the overall global problem. The structure of the model, together with some of the key output data, will be discussed in this paper. It is not claimed that the model, thus far, produces accurate estimates of money laundering flows.

What is defined as a crime in one country may not necessarily be criminal in another. The most profitable crimes in some countries may not be profitable in others. Criminals in some countries might choose to launder their profits, while those in other countries might simply spend them. To this extent, Morris's conclusion that there is no single model that explains money laundering may be correct. However, there may be only a relatively small number of variants of a basic formula. One might be able to say, for example, that "in countries like X, the average profit per recorded fraud is probably around $20,000, but in countries like Y the figure is more like $2,000". Or "in countries like A, around 60% of the proceeds of crime will be laundered, while in countries like B it is likely to be only around 20%".

There is a surprising amount of information about global trends in crime and in money laundering. For example:

• United Nations Crime and Justice databases, describing crimes officially recorded at the national level in over eighty countries;
• International Crime Victims Surveys, that provide insights into the relationships between crime (including crimes not officially recorded) and national socio-economic characteristics in over sixty countries;
• Estimates of the proceeds of crimes – particularly drug-related and other trans-national crimes
• Indices of corruption and susceptibility to money laundering, such as those compiled by Transparency International or the Australian Office of Strategic Crime Assessments in Canberra;
• Geographic, demographic, economic, trade and finance data at the national and international levels.

More is in the pipeline, since the United Nations Centre for International Crime Prevention is currently pilot-testing a survey of trans-national crime, including questions on international linkages between crime groups.

This paper tries to demonstrate that such data can be assembled to produce a model that, while currently lacking some obvious elements, appears to show the way forward. The model, as envisaged in the 1995 AUSTRAC publication that estimated the extent of money laundering in and through Australia, has something of the style of an international input-output model. It proceeds by estimating the quantity of money that could be generated by crime and made available for laundering in each of 226 countries. It then addresses the question of what proportion of this money is likely to be laundered within the same country or sent to another country for laundering, and finally determines which destination countries will receive the funds exported and in what proportions. When this process is complete, the total estimated flows into and out of each of the individual countries can be added up to provide global aggregates, and country profiles can be derived, highlighting where the greatest flows of hot money are, and identifying the key global problem areas.

The Model

To begin with, it needs to be remembered that money laundering is a flow of funds. There is essentially a place where the money is generated, and a place where it is laundered. Even where crime is organised on a transnational basis, the proceeds of crime can be allocated to the countries in which the various victims of crime live. The money may then, of course, be laundered in the same country in which it was generated, or be sent to another country (or other countries) for laundering. It may, furthermore, flow on from its first placement to other countries, and may often return eventually to the originating country so that the offenders can invest their money into legitimate enterprises in their home country.

However, for the purpose of quantifying money laundering, we do not need to follow the money trails beyond the initial point of laundering, because the transactions from that point onwards have all the legitimacy of ordinary monetary flows. In statistical terms, we would be double counting if we followed hot money all the way round its circuitous path from the scene of the crime to the final investment, and counted the same money each time it moved. If $1 million is earned from crime in Australia and sent, say, to a Hong Kong bank for laundering, and from there via Switzerland to the Cayman Islands, from where it is returned "cleansed" to Australia, it is a nonsense to say that these four moves amount to $4 million of money laundering. If a thief sells a stolen bicycle to a second hand retail shop, we do not count another theft when the bicycle is purchased from the shop, and each time it subsequently changes hands, yet this sort of muddled thinking is apparent even in the most influential of reports on money laundering.

In this model, the quantity of money laundering generated in each country is described as dependent principally upon:

• the nature and extent of crime in that country,
• an estimated amount of money laundered per reported crime, for each type of crime, and
• the economic environment in which the crime and the laundering takes place.

A country that does not have a lot of crime, or whose economy does not provide significant profits to criminal enterprises cannot generate large amounts of money to be laundered. In countries with high crime rates and significant criminal proceeds, the potential for money laundering is clearly higher.

The quantity being attracted to each country is described as dependent upon, inter alia:

• the presence or absence of banking secrecy provisions,
• government attitudes to money laundering,
• levels of corruption and regional conflict, and
• geographical, ethnic or trading proximities between the origin and destination countries.

One would expect initial flows of laundered money to favour countries that have secretive banking practices or poor government control over banking. By contrast, subsequent movements of this laundered money may be expected to favour countries with more respectable and controlled, and therefore safer, banking regimes, but as pointed out above, these secondary flows should not concern us. One would also expect money launderers to take advantage of high levels of corruption, if the corrupt behaviour favours their activities, but to avoid those countries in which there are dangerous levels of conflict or where the corruption is of a form that might put their money at risk. One would further expect higher flows of laundered money between places where geographic proximity, or strong trading or community links such as linguistic or ethnic ties simplify business transactions.

With the flexibility and power of modern spreadsheets, it is possible to build in a large number of complex hypotheses such as these, and modify them as new data comes to light. Further develoment of the theories behind the model could result in the creation of a range of new crime-economic indices, leading to a better understanding of the determinants of criminal profitability and the effectiveness of regulatory crime prevention efforts.

Stepwise through the model:

• 1. As a starting point, the United Nations Centre for International Crime Prevention database of recorded crime statistics – the 'UN Survey on Crime Trends and the Operations of Criminal Justice Systems' – contains data on numbers of crimes recorded per year in almost a hundred countries. These relate to the crime categories of Homicide, Assault, Rape, Robbery, Bribery, Embezzlement, Fraud, Burglary, Theft, Drug Possession and Drug Trafficking.
• 2. It is no secret that there are differences in the ways countries classify and count criminal incidents, and that there are significant differences in the extent that police get to know about crimes. But research has also shown how to read between the lines of official crime statistics, by using crime victims surveys of the kind pioneered since 1988 by the Dutch Ministry of Justice and by the United Nations Inter-regional Crime Research Institute in Rome (UNICRI). Enough is known to "see through" major discrepancies in official crime statistics, and make the necessary adjustments. The results presented later in this report do not yet, however, incorporate any such adjustments, as this requires in-depth research because of the large number of countries involved.
• 3. There are, in addition, a number of countries – mostly smaller, less developed countries – for which we have neither official crime statistics nor crime victims surveys. They are mostly, by definition, not major players in the system. Some, however, are regarded as attractive to those seeking to launder money. No country, therefore, can be left out of the model. Using knowledge of the prevailing socio-economic circumstances of each of these countries, per capita crime rates from similar or neighbouring countries can be applied to their demographic data to estimate likely recorded crime figures. The model, at this stage, simply computes average crime rates per capita for each of twelve world regions, and these values are applied to the population figures for all countries without crime data, but there is considerable scope for more considered analysis.

So, at this stage in the process, estimates have been produced for the numbers of crimes recorded by police in each country in each of the eleven crime types. The accuracy and the comparability of these estimates are currently open to question, but in future versions of the model adjustments can be made where sufficient knowledge exists.

The model then proceeds to estimate the total amount of money that is laundered, for each recorded crime in each country. This is not necessarily the same as the average proceeds per crime, although it would be true if all crimes were recorded and if the total amount being laundered from this type of crime were known. Because we acknowledge the fact that not all crimes (particularly in the very important categories of major frauds and drug crimes) are recorded by the police or other authorities, the best way to calculate this figure is by estimating the overall proceeds of crime, for all crimes of this type, and then dividing this figure by the number of crimes recorded.

• 4. The model's starting point for this stage is the crime-specific estimates of money laundering, obtained in the 1995 AUSTRAC report on Australia. The best Australian estimate of total laundered money for each type of crime is divided by the numbers of those types of crimes recorded per year in Australia – to give an average amount of laundered money generated per recorded crime in Australia. Analysis of the Australian report produces the following approximate figures for money laundered per reported crime:
• • 50,000 per recorded fraud offence
  • $100,000 per recorded drug trafficking offence
  • $400 per recorded theft
  • $600 per recorded burglary
  • $1400 per recorded robbery
  • $225 per recorded homicide, and
  • $2 per recorded assault and sexual assault.

  It is worth repeating that these figures are not estimates of the average amount of money laundered per actual crime, but per recorded crime. This inflates the figure considerably, and will differ from country to country depending on the extent to which crimes are recorded by the authorities – a particularly difficult issue to resolve in the cases of drug crimes and frauds. These estimates for Australia so far have very few equivalents from other countries, but similar methods can eventually be used in other countries to broaden the picture.

  The figures, applied to the estimated numbers of crimes recorded in each country (obtainable from the United Nations Crime and Justice databases, op. cit.), result in preliminary estimates of the generation of hot money in each of these other countries.

•  
• 5. The figures initially resulting from step 4 take no account of the differences between countries in the 'profitability' of crime. Two factors are built into the model: - the overall economic situation, as measured by the Gross National Product per capita, and a hypothesised relationship between the level of corruption in a country and the profitability of frauds.
• 
  On the question of the effect of the GNP, it is unreasonable to assume that, other things being equal, poor countries are as likely to generate high levels of criminal proceeds as richer countries. To take account of this, each country's figures from step 4 are factored up or down by data on gross national product per capita. To maintain consistency with the 1995 AUSTRAC report, Australia's GNP per capita is taken as 1.00, and others are pro-rataed accordingly. Benchmarking studies are required to determine the nature of the relationship between GNP per capita and the proceeds of crime – it is quite probable that a linear relationship is not appropriate. For the time being, however, a linear proportionality is assumed in the model. That is, the proceeds per crime in any given country are assumed to be proportional to that country's GNP per capita.

  Addressing the hypothesis that high levels of corruption may increase the amount of money laundered from frauds, even in countries with relatively low GNPs per capita, the Transparency International Corruption Index, transposed to a scale of 1 (low corruption) to 5 (high corruption), is used to factor up the fraud component of money laundering. For example, while low corruption countries use the Australian-based figure of $50,000 per recorded fraud offence, countries with very high levels of corruption, as measured by the T.I. Index, are effectively given a figure of up to five times this dollar amount. Again, this is an area in which significant new research is required.

At this point in the process, steps 1-5 have generated an estimate, for each country in the model, of the total amount of money, generated by crime in that country, and made available for laundering. The next step is to estimate the proportion of this money that will be laundered within the country – the remainder, of course, would be laundered in other countries.

• 6. In the current model, the proportion laundered internally is calculated using the same 1-5 scale of corruption based on the Transparency International index, assuming that countries with high levels of corruption will allow money to be readily laundered in their own economy and thereby reduce the need to launder in foreign countries. The formula incorporated into the model simply assumes that, for each point on this corruption scale, an additional 20% of the money generated from crime is laundered locally. This results in highly corrupt countries (values approaching 5 on the scale) have 80-100% laundered locally, while those with the lowest corruption scores (values only slightly above 1) have only 20-30% laundered locally. Countries without any score on the TI index have been allocated a score equal to the average for their world trade region.
• 
  The assumptions currently used in Step 6 need to be further addressed from a theoretical standpoint. The logic behind the decision to launder locally or launder in a foreign market does not appear to be well known or quantified. Other indicators, such as whether the country has any "suspect transaction" legislation or monitoring agency, would perhaps be appropriate for inclusion in this formula.
•  
• 7. Finally, the model estimates how the foreign-laundered part of the total generated in each country is distributed amongst the over-200 other countries around the world. The current assumption builds in four likely tendencies: -
• • [i] that foreign countries with a tolerant attitude towards money laundering (e.g. those with banking secrecy laws or uncooperative government attitudes towards the prevention of money laundering) will attract a greater proportion of the funds than more vigilant countries,
  • [ii] that high levels of corruption and/or conflict will deter money launderers, because of the risks of losing their funds,
  • [iii] that countries with high levels of GNP/capita will be preferred by money launderers, since it would be easier to 'hide' their transaction, and
  • [iv] that, other things being equal, geographic distance, and linguistic or cultural differences, work as deterrents to money launderers.

  It is interesting to see the results of the first three of these assumptions, as they can be combined to form an 'index of attractiveness' to money launderers. The formula, in algebraic terms is:

  Attractiveness to Money Launderers

  =

  [GNP per capita]*[3*BankSecrecy+GovAttitude+SWIFTmember–3*Conflict–Corruption +15]

  
  

  Where GNP per capita is measured in US$,

  BankSecrecy is a scale from 0 (no secrecy laws) to 5 (bank secrecy laws enforced),

  GovAttitude is a scale from 0 (government anti-laundering) to 4 (tolerant of laundering),

  SWIFTmember is 0 for non-member countries and 1 for members of the SWIFT international fund transfer network,

  Conflict is a scale from 0 (no conflict situation) to 4 (conflict situation exists),

  Corruption is the modified Transparency International index (1=low, 5=high corruption),

  And the constant '15' is included to ensure that all scores are greater than zero.

  The scores on this index, as they result from the assumptions used in the current model, are presented in the following table. It is important to note that a high score on this index does not necessarily reflect poorly on that country's banking regime or government stance regarding money laundering. High scores on the index can be achieved by providing a secure environment for investments generally, as well as by providing a benign environment for money launderers. Bearing in mind that these scores are based on a very simple formula derived from publicly available information and the researcher's own intuition as to the relative importance of the various factors, most of the country rankings appear to be quite logical.

The final step in this process is to incorporate a 'distance deterrence' assumption into the formula to determine how each country's outgoing money laundering is distributed amongst the 225 other countries. The formula used is:
 

The distances between countries were estimated using an feature of the Mapinfo software, identifying the latitudes and longitudes of the approximate population centroids of each country and using simple geometry to calculate the distances between them. The use of the distances squared as a measure of deterrence uses empirically-based regional economic analysis conventions, by which interactions between communities reduce according to the square of the distance between them.

The geographic distance formula should, after further research, be replaced by a more complex "Index of Trading Proximity", using a formula that would include, in addition to the geographic information, data on bilateral trade and finance, currency transaction reporting statistics, cross-border currency movement reporting figures, and on ethnic and linguistic linkages between countries. In addition, more sensitive measures of corruption, conflict and tolerance of money laundering, including perhaps suspicious activity report statistics, need to be developed.

The Results of the Model

The full spreadsheet occupies 22 megabytes of disk space, and is therefore difficult[!!] to include in full in this document. However, it is interesting simply to present some summary results from the matrix – ie the total money laundering generated in each country and the total money laundering attracted to each region and country. The figures generated by the assumptions described above are presented in the tables below. A total of over $US2.8 trillion is obtained for global money laundering, which is within the range of estimates reported by the IMF (op. cit.).

Table 2 and Figure 1 summarise the estimated international flows of laundered money at the global level. Note that, in these figures, flows of money generated and laundered in the same region of the world may actually involve international transfers (e.g. a flow from the U.K. to Switzerland would be included in the internal figure of $985 billion for money generated and laundered in Europe).

Table 2. Estimates of the Major Money Laundering Flows Around the World. ($USbillion/year)  

Figure 1. Estimates of the Major Money Laundering Flows Around the World.

($USbillion/year) 

The model actually produces estimates at the level of individual countries. It is very important to reiterate that these figures represent only an interim set of results and not the author's best and final estimates of money laundering around the world. They are included to show the types of output that would be derived from a fully developed model, and cannot yet be regarded as serious measures of money laundering flows.

Readers may note, for example, that some of the figures of money laundering currently derived by the model amount to rather more than the entire recorded GNP of some countries, and while this may in fact not be impossible, it indicates that, as discussed earlier, the model probably needs to pay more attention to constraints involving actual economic and financial transaction data.

More work is definitely required before the output of this model may be considered to be an adequate response to the question of quantifying global money laundering, but the approach appears to be feasible and capable of further refining.

Table 3 shows the top twenty countries of origin for laundered money, as estimated by the model. Note that most are developed countries.  

The model then tries to estimate where these amounts of hot money will go for laundering, using the assumptions described above. Estimates of the top twenty flows are presented in Table 4, including flows of funds within the generating countries themselves. 

Finally, it is possible to aggregate these flows according to their destinations. Table 5 presents the top twenty destination countries for money laundering, according to the assumptions currently incorporated in the model. 

It is interesting again to note how much of the laundered money, using these assumptions, flows to already developed countries – particularly the United States and Europe. The potential of money laundering to widen the gap between rich countries and poor countries is another important issue that can be tested using a model of this kind.

Use of Media Content Analysis for Calibration of the Model

As a means of evaluating the credibility of the estimates produced by the model, a sample of one hundred press clippings on money laundering or related issues, provided by a crime-related media monitoring service, was examined for information regarding the extent of national or global flows of laundered money.

The original press reports, predominantly (but not exclusively) from English-language printed and electronic media, were dated between 27 February and 5 May 1998 – a period of less than ten weeks. More recently, national assessments for Belarus (personal communication), Canada (web site) and Colombia (clippings) have also been obtained, together with an estimate for drug-related money laundering in the USA (Europol clippings).

Particular passages in the press clippings were extracted, relating specifically to the amounts of money being generated by crime and laundered around the world, examples of types of crime that generate launderable levels of criminal proceeds, the countries in which they take place, and the means by which the money is laundered. Other passages extracted provide information on the degree of effort made by governments to prevent money laundering in each country. An essential element in the selection of these extracts is that they relate to specific countries. Finally, a number of other extracts have a broader focus – providing global or regional estimates of crime or of the extent of money laundering.

Table 6 summarises the key findings from these clippings, together with the equivalent model results. Bearing in mind that there is much that remains to be done in refining the data and relationships built into the model, these results are already interestingly close to the published assessments contained in the press clippings.

Table 6. Comparisons of Estimates contained in Media reports against Model results.
 

The Walker model of global money laundering relies upon a wide range of risk assessment indices, including crime and economic statistics alongside subjective assessments such as Transparency International's well-known Corruption Index. While such information does not provide absolute numbers for estimates of the proceeds of crime and of money laundering, it provides information on the likely limitations on criminal proceeds and on levels of money laundering in a given country.

"Harder" evidence – i.e. data on actual cases with estimates of the monetary amounts involved - is required to ensure the model 'fits' the available data and therefore has overall credibility. The hard data could be compared with the estimates that emerge from the model, and any discrepancies can be used to adjust or calibrate the assumptions of the model. Such official data are, regrettably, extremely rare owing to the complex and covert nature of the money laundering activity itself. Neither is the extent of the profits from crime a statistic readily obtained from the entrepreneurs themselves.

This small collection of press clipping extracts has, however, revealed useful information on a remarkably broad range of countries (84 in all), crime patterns and money laundering techniques. It has revealed a large number of linkages between criminal groups operating across international borders, and it has provided estimates of the dollar values involved in their financial transactions. All of this information can be used to enhance the model's credibility in the fine detail, and hence its overall credibility.

As it stands, it could not yet be described as an entirely rigorous technique for the identification of key data on money laundering. For example, there is likely to be some unevenness in the international coverage, because the service focuses mainly on European or U.S.-based, English-speaking news services. The researcher's own limited linguistic ability further reduced the scope of the analysis to press reports written in English, simple French or the very rare instance of monosyllabic German. Repetition of high-interest cases, such as the Salinas investigation involving Mexico, Switzerland and Colombia, might also appear to introduce biases or even double counting into the analysis.

On the other hand, one should not be too dismissive of a technique that provides information about over eighty countries from a mere ten weeks supply of press clippings. One might therefore conclude that on-going monitoring of this press clipping service could contribute significantly, and without any major research cost, to the analysis of global money laundering flows.

While it might be less than completely satisfying to evaluate an economic model through its success in predicting expert assessments, rather than through its performance in predicting actual economic statistics, one might be excused on the grounds of the peculiar nature of the crime economy and the complexity of the laundering processes that facilitate it.

Conclusions

This paper has presented the design of a model for estimating flows of money laundering around the world. While there are many problems with missing and non-comparable data, there also appear to be rational techniques for using expert knowledge to fill in these gaps. The model concentrates on assembling or estimating information that can be cross-checked, so that while it will, inevitably, be in error in some areas due to poor data or incorrect hypotheses, there are numerous opportunities to cross-check with other data in the model. For example, estimates based on data and hypotheses about crime levels and profits logically cannot be in conflict with estimates based on economic or financial data. Also a number of ratios and indices (e.g. money laundering as a percentage of GNP, the ML Attractiveness Index) are calculated for every country within the model that can be assessed by expert opinion. Whenever they are in conflict in the model, this is a signal that a 'third opinion' is required – i.e. more research needs to be done in precisely the area of data conflict.

Areas identified in this paper for further research include:

• The estimation of crime levels in countries for which no statistics exist, by the use of demographic and socio-economic data that are more readily available.
• Estimating the relative amounts of money laundered per recorded crime, in each crime type, in a range of country types (e.g. development level, transitional, geographic region etc), and the relationships between these amounts and national indicators such as Gross National Product per capita and the types and levels of corruption.
• Research into the factors determining the decision of where to launder the proceeds of crime; i.e. the proportion of money that is laundered in the country in which it is generated, and the relative attraction of foreign destinations.

The other necessary ingredient is only that the world's great organisations and interest groups combine to enable the necessary research.