I NTERPRETATION

One output (TBML risk) is provided from the presented tables in Figure 12, this output has five different possible outcomes: very low (VL), low (L), medium (M), high (H) or very high (VH). There are a total of five possible ways to reach the outcome

(VL) and two, to reach outcome (L). These two are excluded as ‘least concern’, as they have a very high possibility of providing false positives, as they contain little to no evidence of TBML. As previously noted by Khanna (2016, 7), false positives are an important challenge to take into account in detection. From this conclusion the following type of rule can be derived:

RULE 1.

IF TBML risk is very low OR TBML risk is low

THEN Case status: least concern

The left over outcomes are assigned priority values, such that: medium (M) is ‘P3’, high (H) is ‘P2’ and very high (VH) is ‘P1’. ‘P1’ being of highest priority case and should definitely be checked further, while ‘P3’ being of lesser priority, but should also be checked regardless. As discussed before, the interpretation of fuzzy rules follows the idea that for the outcome, or consequent, the maximum degree of fulfillment of the antecedent will be considered the membership degree.

To grasp an idea of how the presented rule base performs. Dummy input data, previously stated, are inserted and the output observed. If a case were to be handled by a screening expert, they would determine their values for each of our presented variables. As mentioned before, linguistic variables, linguistic values and their ranges should be chosen by an expert or derived from data, if data is available. Therefore, the expert knows exactly what each variable, value and range represents.

The presented data corresponds in our model in the following way: crisp input 0.9 for volume_of_trade corresponds to the membership function high to the degree of 1, country_of_destination input 0.5 corresponds to membership function medium risk to the degree of 1 and price_overvaluation input of 0.8 to the membership ship function high to the degree of 1. If the input corresponds to more than one membership function, then depending on the rule base, the operations union (OR) and intersection (AND) are used to acquire a single value. For instance, the crisp input 0.25 for country_of_destination would map to membership functions low risk and medium risk at degrees of: 0.25 MR and 0.5 LR. This would require intersection or union operations depending on the antecedent. For our dummy data, the outcome of the scenario is formed from Figure 12. and presented in the form of the following rule:

RULE 2.

IF volume of trade is high

AND country of destination is medium risk AND price overvaluation is high

THEN TBML risk is very high;

case status: definitely check - P1

The rule base presents numerous options for the same outcome, for instance, if price_overvaluation is high and just one of the two other variables is above the lowest value, the outcome is very high. It is possible to simplify our rules by the use of the union operation. Let ‘above low risk’ correspond to ‘MR or HR’ and ‘above low’

correspond to ‘L or H’, thus forming the following rule:

RULE 2.2.

IF price overvaluation is high

AND country of destination is above low risk AND volume of trade is above low

THEN TBML risk is very high

If it were the case that no data can be gathered regarding a variable, for instance, in the previous case there was no information about the price of the invoice regarding the shipment. This results in a situation, where the possible outcomes are: low if price_overvaluation is low, high is price_overvaluation is medium, very high if price_overvaluation is high. The suggestion being, as two of the possibilities are P2 and P1 in priority, that the worst case is assumed. To further back this decision, investigation. A set of known TBML cases would be of most value. It would provide validation that the presented model in fact is capable of flagging suspicious cases as priority for further inspection, thus possibly resulting in newly detected cases, proving the value of such a system. In the case of showing, that the model is not capable of accurately flagging cases, further tuning of the linguistic variables, linguistic values and their ranges can be made until success is achieved. As the model is only a partial and mostly theoretical framework, it is suspected that data would provide proof,

that much is needed in terms of tuning. Furthermore, it would provide the opportunity to expand, in the form of additional variables and rules.

5. Conclusions and future research

This study provided an overview of the phenomenon: trade-based money laundering, covering methods, red flag indicators and a literature review. It introduced theory on fuzzy sets, fuzzy numbers, basic operations on fuzzy sets and linguistic variables.

Lastly, these two spheres of study were merged into a prototypical framework of a decision support system, including a rule base, which was constructed using an example case study of TBML.

The research question of this study was: “How to build a decision support system based on linguistic fuzzy modelling for screening of trade transactions for trade-based money laundering schemes”. The question was answered through building the basis of a rule-based model. It became clear, that TBML as a research subject, and as a world phenomenon, is an intriguing challenging one to dissect. Data is difficult to obtain, as such knowledge is best not given out to the public, thus consequently also criminals to study and learn how to divert AML efforts. TBML as a world phenomenon is hazy. Jurisdictions show a lack of awareness and understanding of this particular type of money laundering. These aspects made the research question an interesting one. As far as to our knowledge, linguistic fuzzy modelling has not been applied to a decision support system in this particular field. The inherent fuzziness of forces affecting TBML detection makes such an application viable.

While this study was by no means an exhaustive take on possible aspects of building a comprehensive application, it provided the theoretical basis which is necessary to

proceed to more advanced venture, along with examples on how these ideas can be used in further research. The process of acquiring live data is still ongoing, though looking promising. When and if this happens, it will be a very interesting opportunity to begin working on a more advanced model in the form of software. Real world data could disclose new aspects and possibilities that we may have not foreseen. Tuning the model to suit the needs of detection specialists seems an excitingly challenging endeavor.

Automating the system by coding it into software is an obvious future research idea.

A platform should be formed on which to build a larger rule base and tuned to fit the needs of screening officials in different parts of the world. It would also be interesting to study how such a system translates to different languages and how cultural differences play a part in modifying the meanings of linguistic variables used.

The representation of model results in graphical term seems an interesting aspect to look into. Depending on how the graphical output is laid out, the effects on the user may present opportunities and challenges. In detecting cases it may be very important to acquire results quickly as to not hinder the speed of work, this will require extra attention to graphical output. This aspect of modelling was quite limited in this particular study, but should see more attention in future research.

Reference list

APG. 2012. APG Typology Report on Trade Based Money Laundering. Asia Pacific Group. [26.1.2017]. From:

http://www.fatf-gafi.org/media/fatf/documents/reports/Trade_Based_ML_APGReport.pdf

Arfi, B. 2010. Studies in Fuzziness and Soft Computing, Volume 253. Springer.

Bhagwati, J. 1964. On the underinvoicing of imports. Bulletin of the Oxford University Institute of Economics & Statistics. Vol 27, Issue 4, 289-397.

Brown. 2009. Free trade zones: Haven for money laundering and terrorist financing?

ACAMS Today. January: 10-12

De Boyrie, M.E., Pak*, S.J. and Zdanowicz, J.S., 2005. The impact of Switzerland's money laundering law on capital flows through abnormal pricing in international trade. Applied Financial Economics, 15(4), pp.217-230.

Delston, R.S. and Walls, S.C., 2009. Reaching beyond banks: How to target trade-based money laundering and terrorist financing outside the financial sector. Case W.

Res. J. Int'l L., 41, p.85. [19.2.2017]. From:

http://scholarlycommons.law.case.edu/jil/vol41/iss1/5

EAG. 2009. International trade based money laundering. The Eurasian Group.

[15.2.2017]. From:

http://www.eurasiangroup.org/restricted/WGTYP_report_3_2009_62_eng.doc

EAG. 2010. Risks of Money Laundering in Foreign Trade Transactions. The Eurasian Group. [15.2.2017]. From:

http://www.eurasiangroup.org/ru/news/WGTYP_2010_6_eng.pdf

FATF. 2006. Trade-based money laundering. Financial Action Task Force (FATF).

[4.2.2017]. From:

http://www.fatf- gafi.org/publications/methodsandtrends/documents/trade-basedmoneylaundering.html

FATF. 2008. Best practices paper on trade based money laundering. Financial Action Task Force (FATF). [4.2.2017]. From:

http://www.fatf-gafi.org/documents/documents/bestpracticesontradebasedmoneylaundering.html

FATF. 2010. Money laundering vulnerabilities of free trade zones. Financial Action Task Force (FATF). [1.2.2017]. From:

http://www.fatf-gafi.org/publications/methodsandtrends/documents/moneylaunderingvulnerabilitiesoff reetradezones.html

Ferwerda, J., Kattenberg, M., Chang, H.H., Unger, B., Groot, L. and Bikker, J.A., 2013. Gravity models of trade-based money laundering. Applied Economics, 45(22), pp.3170-3182.

FinCEN. 1997. FinCEN Advisory: Colombian Black Market Peso Exchange. United States Department of Treasury Financial Crimes Enforcement Network (FinCEN).

Issue 9. [28.4.2017]. From:

https://www.fincen.gov/sites/default/files/shared/advisu9.pdf

HKAB. 2016. Guidance paper on combating trade-based money laundering. Hong Kong Monetary Authority. [16.3.2017]. From:

http://www.hkma.gov.hk/media/eng/doc/key-functions/banking-stability/aml-cft/Guidance_Paper_on_Combating_Trade-based_Money_Laundering.pdf

Khanna, M. 2016. Trade based money laundering - Capturing the new frontier through analytics. CAMS AUDIT White Paper. ACAMS. [14.3.2017]. From:

http://files.acams.org/pdfs/2016/Trade_Based_Money_Laundering_Capturing_M_Kh anna.pdf

Klir, G. Yuan, B., 1995. Fuzzy sets and fuzzy logic: theory and applications. New Jersey, Prentice Hall.

Liao, J. Acharya, A. 2011. Trans-shipment and trade-based money laundering.

Journal of Money Laundering Control. 14(1): 79-92.

McDowell, J. and Novis, G., 2001. The consequences of money laundering and financial crime. Economic Perspectives, 6(2), pp.6-10.

McSkimming, S., 2010. Trade-based money laundering: Responding to an emerging threat. Deakin Law Review. Vol 15, No 1.

Negnevitsky, M., 2005. Artificial intelligence: A guide to intelligent systems. Biddles Ltd, King’s Lynn. Great Britain.

Plemenos, D., & Miaoulis, G. 2008. Artificial Intelligence Techniques for Computer Graphics. Springer Berlin Heidelberg.

RJ Soudijn, M., 2014. A critical approach to trade-based money laundering. Journal of Money Laundering Control, 17(2), pp.230-242.

Stoklasa, J., 2014. Linguistic models for decision support. Dissertation. Acta Universitatis Lappeenrantaensis 604.

The Wolfsberg Group. 2017. Trade Finance Principles. The Wolfsberg Group, ICC, BAFT. [4.4.2017]. From: http://www.wolfsberg-principles.com/pdf/home/Trade-Finance-Principles-Wolfsberg-Group-ICC-and-the-BAFT-2017.pdf

NMLRA. 2015. National Money Laundering Risk Assesment. Department of the Treasury’s Office of Terrorist Financing and

Financial Crimes (USA). [02.05.2017]. From: https://www.treasury.gov/resource-

center/terrorist-illicit-finance/Documents/National%20Money%20Laundering%20Risk%20Assessment%2 0%E2%80%93%2006-12-2015.pdf

Walker, J. and Unger, B., 2009. Measuring global money laundering: the Walker gravity model. Review of Law and Economics, 5(2), pp.821-853.

WTO. 2016. World trade statistical review 2016. World Trade Organization (WTO).

3:18. [10.1.2017] From:

https://www.wto.org/english/res_e/statis_e/wts2016_e/WTO_Chapter_03_e.pdf

Zadeh, L.A. 1965. Fuzzy sets. Information and control. Vol 8, Issue 3, p. 338-353.

Zadeh, L.A., 1975. The concept of linguistic variable and its application to approximate reasoning-I. Information sciences. Vol 8, Issue 3, p. 199-249.

Zdanowicz, J., 2004. Who’s watching our back door? Business Accents. Florida International University. Vol 1, No. 1, 26. [10.1.2017]. From:

https://business.fiu.edu/pdf/publications/business-accents_2004.pdf

Zdanowicz, J., 2009. Trade-based money laundering and terrorist financing. Review of Law and Economics. 5(2): 858-878.

In document Linguistic fuzzy modelling: support for detecting trade-based money laundering (sivua 42-0)