10-15-2008b-Wed_MHTM-UNCOOKING THE BOOKS FROM TOX=
IC
PAPER SUB-PRIME MORTGAGES CDS AND CSOS MATERIAL MISSTATEMENTS.mht
UNCOOKING THE BOOKS FROM TOXIC PAPER SUB-PRIME MORTGAG= ES CDS AND CSOS MATERIAL MISSTATEMENTS OF THE FINANCIAL SERVICES INDUSTRY: A BARCLAYS BANK PLC CASE STUDY
Avi
Rushinek and Sara Rushinek, University of Miami, arush@miami.edu or arushine@aol.com
(305)666-7890 421 Jenkins Bldg, Coral Gables, Fl 33124, USA
Abstract
This is a study of the sub-prime mortgages, Credit Def=
ault
Swaps (CDS), and Collateralized Synthetic Obligations (CSOs) cooking the bo=
oks
of the financial services industries around the world. This case study develops a methodo=
logy
of uncooking the books from material misstatements of the financial industry
using Barclays Bank as a classic case study. It shows how sub-prime mortga=
ges, CDS,
and CSOs overstated the revenues of the financial services industries leadi=
ng
to the stock markets melt-down of October 2008. This research develops multiple
regression models and software that simulates such frauds automatically.
This
study develops a method for fraud “finger print” definition. The
regression parameters define parts of the fraud finger print. Some of these
parameters include such variables as the upper and lower bounds of the
coefficients of the best fraud predictor variables, and its statistical
significance. These finger prints define Expert System (ES) rules for a Case
Based Reasoning (CBR) Knowledge Base (KB). The CBRKB should eventually cont=
ain
all the possible combinations of fraud finger prints for each Standard Indu=
stry
Classification (SIC) code. We also define the fraud rate of over or under
statements of financial accounts, and its time periods, as well as its amou=
nts
and their pattern, since they may affect the fraud finger print. Such finger
prints together with other fraud characteristics should help an ES diagnose
fraud patterns by benchmarking a suspicious company against its peers. The
resulting anomalies will fire the ES rules that will help trace the source =
of
the fraud.
* We omitted additional appendices, tables,
glossaries, program code listings and screen shots to save space. We will
provide such materials upon written request.
Introduction & Literature Review
BARCLAYS BANK COOKING THE BOOKS MATERIAL MISSTATEMENTS GOOGLE SEARCH PHRASE
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
A search of Google using the phrase "Barclays Bank
cooking The Books material misstatements" reveals the following hyperl=
inks
among others: Peter Viles, peter.viles@latimes.com, May 2007, writes on the=
http://latimesbl=
ogs.latimes.com/laland/mortgages/ about "Blame game: KPMG accus=
ed of
lax auditing of New Century."
The internet sites report that "Driven by a 'brazen obsession' =
with
generating sub-prime mortgages,
"Michael J. Missal's report said senior managers 'turned a blind eye' to the 'ticking time bomb' created by the high-risk lending in 2005 and 2006. At the same time, Missal said, New Century's audi= tor, KPMG, contributed to the problems by failing to exercise due care in review= ing its books, leading to material misstatements in New Century's financial reports."
From The New York Times: "In a sweeping accusation against one of the country’s largest accounting firm, an investigator released a report on Wednesday that said 'improper and imprudent practices'= by a once high-flying mortgage company were condoned and enabled by its audito= rs."
"KPMG denies the accusations in the report, which= was commissioned by the United States Trustee overseeing the New Century bankruptcy."
BARCLAYS GROUP
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
The same site reports about "Ranking lenders by r=
isky
loans." It asks: "Why are some lenders failing and others hanging
in?" Ranking agencies like "SMR ... ranked 163
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D
CREDIT DEFAULT SWAP CDS
Matthew Monks, September 9, 2008 2:28 PM ET, on (Matth= ew Monks at mmonks@financialweek.com and the editors at fw_editor@financialweek.com) http://www.financialweek.com/apps/pbcs.dll/article?AID= =3D/20080909/REG/809099979/1036 reports that "Mortgage b= ailout could trigger massive credit default swap settlement Payouts to be limited,= though, as Fannie, Freddie bonds likely to be settled close to par."
“The government takeover of Fannie Mae and Fredd= ie Mac could trigger the largest credit default swap settlement ever. Actual payme= nts could be limited, however, as a result of the relatively high value of the mortgage underwriters’ bonds. Investors may be forced to settle contr= acts covering the mortgage giants’ $1.6 trillion in outstanding debt becau= se the government seizure constitutes a credit event that triggers the payment= or delivery of their bonds. The International Swaps and Derivatives Association announced on Monday that it would establish a protocol to facilitate the settlement of CDS trades involving Fannie Mae and Freddie Mac.”
“This will likely be the largest CDS credit even= t in terms of the amount of CDS contracts outstanding,” J.P. Morgan analyst Eric Beinstein wrote in a report released on Mo= nday "The settlement process will likely take 30 days, with investors cashi= ng their CDS contracts at a price established through an auction process. Payo= uts could be limited, though, because most analysts believe CDS covering the mortgage companies’ senior debt will be settled between $98 and par, = RBC Capital Markets analyst T.J. Marta wrote in a report to clients. In a cash = CDS settlement, buyers are paid the difference between the par value and market value of the debt obligation. In related news, at least two analysts believe new Fannie = Mae and Freddie Mac CDS contracts will likely be introduced under different ter= ms.
Barclays Capital analysts Vince B= reitenbach and Jeff Meli wrote in a report that Barclays <= /p>
“believes the appointment of a conservator for t= he [firms] constitutes a credit event for
[their] senior and subordinated CD= S. As such, new CDS contracts without the conservatorship trigger should begin= to trade.”
&&&&&&&&&&&= ;&&&&&&&&&&&&&&&= ;&&&&&&&&&&&&&&&= ;&&&&&&&&&&&&&&&= ;&&&&&&&&&&&&&&&= ;&&&&&&&&&&&&&&&= ;&&&&&
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
FANNIE, FREDDIE CREDIT-DEFAULT SWAPS MAY BE SETTLED
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D
Oliver Biggadike and Shannon D. Harrington, on Sept. 8 (Bloomberg) writes -- "Investors may be forced to settle contracts protecting more than $1.4 trillion of Fannie Mae and Freddie Mac bonds agai= nst default after the U.S. seized control of the companies in a bid to bolster = the housing market (Oliver Biggadike in Sydney at obiggadike@bloomberg.net; Sha= nnon D. Harrington in New York at sharrington6@bloomberg.net).
Thirteen ``major'' dealers of credit-default swaps agr= eed ``unanimously'' that the rescue
constitutes a credit event triggering payment or deliv= ery of the companies' bonds, the
International Swaps and Derivatives Association said i= n a memo obtained by Bloomberg News today.
Market makers for the privately traded contracts will
discuss how to settle them in a conference call at 11 a.m. in
A settlement likely would be the largest in the market= 's decade-long history. Credit-default swaps on Fannie and Freddie have been a= mong the most actively traded the past few months, according to reports from bro= ker GFI Group Inc. Both companies also are among 125 companies in the benchmark= Markit CDX North America Investment Grade Index, the = most actively traded contract in credit markets, which investors use to speculat= e on corporate creditworthiness or to hedge against losses."
CONSERVATORSHIP IS A CREDIT EVENT BARCLAYS PLC ANALYST= S NOTE TO CLIENTS
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
``We believe conservatorship is a credit event,'' Barc=
lays
Plc analysts Vince Breitenbach and Jeff Meli said in a note to clients yesterday. Barclays is=
a
member of the ISDA.
sharring= ton6@bloomberg.net ;
&&&&&&&&&&&= ;&&&&&&&&&&&&&&&= ;&&&&&&&&&&&&&&&= ;&&&&&&&&&&&&&&&= ;&&&&&&&&&&&&&&&= ;&&&&&&&&&&&&
http://www.bloomberg.com/apps/news?pid=3D20601087&a= mp;sid=3Daa6nmsv7BakE&refer=3Dhome claims that "Lehman Credit-Swap Auction Sets Payout of 91.38 Cents&quo= t; by Shannon D. Harrington and Neil Unmack on Oct= . 10 (Bloomberg) -- "Sellers of credit-default protection on bankrupt Lehman Brothers Holdings Inc. will have to pay 91.375 cents on the dollar to settle the contracts, setting up the biggest-ever payout in the $55 trillion marke= t.
An auction to determine the size of the settlement on =
Lehman
credit-default swaps set a value of 8.625 cents on the dollar for the debt,
according to Creditfixings.com, a Web site run by auction administrators
“No one knows exactly who has what at stake beca= use there's no central exchange or system for reporting trades. Sellers are required to post collateral, or pledge assets, to the buyer of protection, known as the counterparty, on the other side of the trade if the value of t= heir positions declines. Because Lehman's bonds had already fallen, the collater= al has probably been posted, Yelvington said."= ;
“Hedge funds, insurance companies and banks typi= cally buy and sell credit protection, which is used either to insure a bond again= st default or as a bet against the company's ability to pay its debt. The paym= ents ``are insignificant when put into the context of the trillions of dollars of payments that are made through settlement systems each and every day.''
“Some funds may be forced to dump assets to meet= the payment demands if they haven't hedged, … Banks can go to the Federal Reserve, or use the commercial paper market where it is still functioning''= to meet protection payments, said Cicione, who sai= d a 9.75 cent recovery rate would lead to payments of about $270 billion. But f= und managers or hedge funds, once they've used their cash, have only one option= : to sell assets.''
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
OFFSHORE BANK, A BERMUDA-BASED COMPANY HAD MADE BETS W= HICH THE ICELANDIC GOVERNMENT SEIZED ... STOCK WAS HALTED FROM TRADING ON THE NEW YORK STOCK EXCHANGE YESTERDAY AFTER FALLING DOWN 89 % PERCENT THIS YEAR ...= THE ICELANDIC BANKS THAT FAILED THIS WEEK WERE ALSO OFTEN INCLUDED IN CDOS CREA= TED DURING 2006
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
A unit of Primus Guaranty Ltd., a Bermuda-based compan= y that has sold more than $24 billion in credit-default swaps, said last month it guaranteed $80 million of Lehman debt. The firm sold protection on $215 mil= lion of Fannie and Freddie debt and $16.1 million on WaMu. Yesterday, it said it also had made bets of $68.2 million on Kaupthing Bank hf, which = the Icelandic government seized.
Primus said last week it had $820 million in cash and = liquid investments to meet claims on the contracts. The stock was halted from trad= ing on the New York Stock Exchange yesterday after falling to 99 cents. The sha= res, down 89 percent this year, slumped 15 cents, or 17 percent, to 75 cents.
The failures of Lehman, once the fourth-largest securi=
ties
firm, and Seattle-based Washington Mutual Inc. as well as the government
takeovers of Fannie Mae, Freddie Mac and
The use of credit derivatives has grown more than 100-=
fold
in the past seven years as investors began using the swaps to bet on compan=
ies'
creditworthiness. The Federal
Reserve Bank of
expedite efforts for a market clearinghouse that would reduce risks and absorb counterparty losses resulting from the failure of market makers such as Lehman."
"``CDS contracts did not cause any firm to fail,'= ' Pickel said. ``The underlying cause of
problems that has affected firms is the risk that they= chose to take on. ''Credit-default swaps are financial instruments that can be ba= sed on bonds and loans. They pay the buyer face value in exchange for the underlying securities or the cash equivalent should a borrower fail to adhe= re to its debt agreements.
Five-year credit-default swaps on Lehman rose as high = as 790 basis points before the firm filed for bankruptcy, according to Phoenix Partners Group, a New York-based inter-dealer broker. A basis point on a credit-default swap contract protecting $10 million of debt from default fo= r five years is equivalent to $1,000 a year.
Standard & Poor's has rankings on 1,889 CDOs that = sold credit-default swap protection on Lehman, the New York-based ratings firm s= aid last month. Pieces of 1,526 CDOs sold protection on Washington Mutual, S&am= p;P said. More than 1,200 made bets on both Fannie and Freddie. The Icelandic b= anks that failed this week were also often included in CDOs created during 2006 = and 2007, according to Sivan Mahadevan, a New York-= based Morgan Stanley strategist."
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
AS LAW PROFESSOR WARN THE INTERNET ABOUT TOXIC PAPER R= ISK THE AUDITORS FIND NO MATERIAL MISSTATEMENTS
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
As Law Professor Warn The Internet About Toxic Paper R= isk And Material Misstatements
------------------------------------------------------= ------------------------------
http://lawprofessors.typepad.com/securities/news_stori= es/index.html, reports on
August 23, 2007 "Rumors of a
On March 27, the site reports "New Century's Bank= ruptcy Expected," ... the poster child for the collapse of the subprime mortgage industry, is expected to fi= le for bankruptcy soon, as both Barclays Bank ... took back loans that secured= New Century's financing and plan to auction them off.
On September 22, 2008 the site writes about "Bank=
ruptcy
Court Approves Sales of Lehman Broker Dealer to Barclays." They procee=
d to
report: "Questions raised about Lehman's transfer of assets among its
businesses in the days before its bankruptcy filing did not interfere with
swift action by the bankruptcy court to approve the sale of substantially a=
ll
of the assets of Lehman Brothers, Inc., to Barclays Capital. The court̵=
7;s
decision followed a marathon eleven-hour hearing in a packed
new firm. The expeditious t= ransfer of Lehman’s assets also avoids disruption of capital markets because securities transactions will continue to be completed and Lehman’s counterparties can confidently continue to do business with the firm."=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
THE AUDITOR REPORT THAT THE FINANCIAL STATEMENTS ARE F= REE FROM MATERIAL MISSTATEMENTS
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
&&&&&&&&&&&= ;&&&&&&&&&&&&&&&= ;&&&&&&&&&&&&&&&= ;&&&&&&&&&&&&&&&= ;&&&&&&&&&
ht= tp://www.secinfo.com/d14D5a.t2zzz.htm reports about "CDO Credit Default Swap Barclays Bank settlement CBOE Misstatement." Quoting from the "Report of I= ndependent Registered Public Accounting Firm To the Board of Directors and Shareholder= s of Barclays Bank PLC," the accountant’s state:
In our opinion, financial statements are free of $$ ma= terial misstatement." PricewaterhouseCoopers LLP, Chartered Accountants and Registered Auditors, London, United Kingdom, 10th March 2008, proceed in the "Internal control" section of Barclays report they repeat the assertion that they provide " reasonable assurance against $$ material misstatement or loss."
In page 148 of Barclays, Annual Report 2007, the audit= ors use the same boiler plate stating that: "Internal control systems obta= in reasonable assurance about whether the financial statements are free of material $$ misstatement ..."
IMPLYING THAT CDOs ARE SAFE ASSETS, AND UNDERSTATING T= HE TOXIC INHERENT RISK.
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
In the "Corporate bonds" section they report states that "Corporate bonds are generally valued using observable quo= ted prices or recently executed transactions. Where observable price quotations= are not available, the fair value is determined based on cash flow models where significant inputs may include yield curves, bond or single name credit def= ault ^^ swap spreads." They n= ever mention any of the reports on the internet concerning the toxic paper, and = the debate in the press.
in the "Derivatives" section the report talk= s some more about the swaps explaining that
"Derivative contracts can be exchange traded or o= ver the counter (OTC). OTC derivative contracts include forward, ^^ swap and op= tion contracts related to interest rates, bonds, foreign currencies, credit stan= ding of reference entities, equity prices, fund levels, commodity prices or indi= ces on these assets.
For many pricing models there is no material subjectiv= ity because the methodologies employed do not necessitate significant judgment = and the pricing inputs are observed from actively quoted markets, as is the case for generic interest rate ^^ swaps and option markets. In the case of more established derivative products, the pricing models used are widely accepted and used by the other market participants," implying that they are safe assets, and understating the toxic inherent risk.
^^ swap spreads - indicate discussion about swap sprea= ds and CDOS
## CDO Super Senior - indicate discussion about swap s= preads and CDOS
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
EXPOSURES RELATING TO US SUB-PRIME WERE ACTIVELY MANAG= ED AND DECLINED OVER THE PERIOD REPORT CLAIMS
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
The report proceeds to understate the risk of sub-prime toxic paper, stating that "Undrawn contractually committed facilities = and guarantees provided includes £360m (2006: £nil) provision again= st undrawn facilities on ABS ## CDO Super Senior positions.
The
Impairment charges and other credit provisions of £846m included £722m against ABS ## CDO Super Senior exposures, £60m from other credit market exposures and £58m relating to dr= awn leveraged finance underwriting positions. Other impairment charges on loans= and advances amounted to a release of £7m (2006: £44m release) befo= re impairment charges on available for sale assets of £13m (2006: £86m)."
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
MISREPRESENT REALITY BY APPLYING "
UNDERSTATE THE TOXIC PAPER TRUE RISK EXPOSURE
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
http://msnmoney.bra= nd.edgar-online.com/EFX_dll/EDGARpro.dll?FetchFilingHTML1?ID=3D5823375&= SessionID=3D5RgcWZDBP11rCl9 in the "– Structured cre= dit derivatives" section of the report states that:
"Collateralised synth= etic obligations (CSOs) are structured credit derivatives which reference the lo= ss profile of a portfolio of loans, debts or synthetic un= derlyings. The reference asset can be a corporate credit or an asset backed credit. For CSOs that reference corporate credits an analytical model is used. For CSOs= on asset backed underlyings, due to the path depen= dent nature of a CSO on an amortizing portfolio a Monte Carlo simulation is used rather than analytic approximation. The expected loss probability for each reference credit in the portfolio is derived from the single name credit default ^^ swap spread curve and in addition, for ABS references, a prepaym= ent rate assumption.
A simulation is then used to compute survival time whi=
ch
allows us to calculate the marginal loss over each payment period by refere=
nce
to estimated recovery rates. Significant inputs include prepayment rates,
cumulative default rates, and recovery rates." Again, they misrepresent reality by
applying "
NET
SALES/TOTAL ASSETS ESTIMATING 10% INFLATED NET SALES & RECEIVA=
BLE,
TO ENLARGE GROSS PROFIT & STOCK MARKET VALUES: A FBARCLAYS BANK PLC
AND THE COMMERCIAL BANKS, NEC
(COB) SIMULATED FRAUD
SOFTWARE CASE STUDY
Overview
This study finds the ratios NET SALES/TOTAL ASSETS, SG
& A/SALES and <=
u>NET SALES/EMPLOYEES<=
/span> as the Best Predictor Percent Differenced
Financial Ratios (BP%DFR) and most effective in constructing a fraud cost
model. These BP%DFR variables are estimating 10%
INFLATED NET SALES & RECEIVABLE, TO ENLARGE GROSS PROFIT & STOCK MA=
RKET
VALUES fraud value. We use=
these
variables to build a regression model. This is a case study of
We Simulated Fraud (SF) o= f the NET SALES, which is the primary (1st) acco= unt. We balance this 1st account against the RECEIVABLES account, the 2nd account. The FF remains fixed starting from the= 1st period, 12/31/85, to the last peri= od, 12/3= 1/94, at $1,318,800. We have calculated = the FF as the higher of 10% of the 1st primary SF account or 1% of Net Sales. Add the FF to the VF a= nd you get the MF for the last year $2,637,600. Add this MF to the initial balance= of the 1st account, the real account balance. This rNET= SALES balances $13,188,000, for the last period of this study. This way, you get the phony (f=3Dfraudulent) balance = of 1st account, fNET SALES, $15,825,600. Likewise, for the 2nd balancing account, start with the real balance, (r prefixed account), $115,356,000, combine the SF, $2,637,600, and you get the phony value (f prefixed) of the fRECEIVABLES bala= ncing the SF at $117,993,600.
We difference (deduct) th=
e fNET SALES (Fraudulent, Post SF) from the,=
rNET SALES (real, Pre SF), calculating the =
SF
Dollar values. Likewise, we difference all the Disclosure™ financial
ratios (%DFR), to identify the fraud drivers. We regress the SF Dollar Valu=
es
on the %DFR, to discover the Best (hi=
ghest
R Square) Predictor Percent Differenced Financial Ratios (BP%DFR). In this
case, these (BP%DFR) include: NET
SALES/TOTAL ASSETS, SG =
&
A/SALES & NET SALES/EMPLOYEES. &n=
bsp;
INTRODUCTION OF PERCENTAGE DIFFERENCED FINA=
NCIAL
RATIOS (%DFR)
According to new securities legislation, au=
ditors
will need to report more quickly than at present any suspicion of fraud. A
tougher approach is also coming from the auditing standards board of the AI=
CPA.
The new standard will require external auditors to be more aggressive in not
only reporting but unearthing fraud and faulty financial statements. Failur=
e to
comply with the institute’s measure will include civil penalties and
possible loss of a CPA’s license. These proposed standards will have
auditors spotting higher risk of fraud to develop specific plans to elimina=
te
that risk. The board has been considering stiffening auditing standards for
several years as research showed that current standards were not tough enou=
gh
on discovering major frauds (Berton, 1996).
This study defines the %DFR as a predictor
variable for estimating the value of fraud. We are developing several theor=
ies
and applying other theories to design this study. The Reversed Accounting
Theory (RAT) contends that to discover fraud we have to go from the final
financial statements, in reverse, to the original fraudulent transaction. To
pick the proper methodology, we are applying Activity Based Costing (ABC) to
RAT and fraud, developing a derivation of ABC, Activity Based Fraud (ABF).<=
span
style=3D'mso-spacerun:yes'> ABF contents that like any other c=
ost,
fraud originates from some activities. Since we do not know what these
activities are, we could use highly correlated surrogates to these activiti=
es
to act as fraud drivers. These %DFR are such drivers.
We
intend to supplement existing Expert Systems (ES) software that can
discriminate between fraudulent and fraud free companies, but cannot pinpoi=
nt
the amounts and the accounts. Integrating
such a model into these ES, will extend their abilities beyond current
technology. Such extension will enable the ES to quantify the fraud and flag
its sources (accounts). The supplementary nature of model we develop explai=
ns its
limitations. This model is not for discovering fraudulent companies, since =
it
will be redundant to the existing ES. Thus, we optimize it for a relevant r=
ange
that outside the vicinity of the origin, where both the %DFR and the Fraud
approach a value of zero. Therefore, we hypothesize that the intercept will=
be
statistically insignificant about the origin, which it turns out to be. In
contrast to the insignificance of the Intercept, we hypothesize that the en=
tire
model will be significant, as well as at least one PERCENT. Indeed, the ANO=
VA
(Analysis Of Variance) confirms our expectations, rejecting the null hypoth=
esis
that the model is insignificant and that the regression coefficients are eq=
ual
to zero.
Since we intend to use it ultimately on a fraud suspect (a company the E has identified) we refer to = that company as the Focus Company. In contrast, the industry or competitors̵= 7; ratio averages constitute a fraud free Peer Review Group (PRG). We calculate the percentage as the ratios of the Focus less the Peers divided by the Pee= rs. These assumptions that fraud is the single difference between the Focus and= the Peers highlight the limitations of this approach. These limitations exclude heterogeneous industries where companies are not very similar, as well as countries where fraud (e.g., bribes) is part of doing business, and GAAP/S (Generally Accepted Accounting Principles and Auditing Standards) are rare.=
We search for the ratios that will maximize=
the
correlation between the %DFR, the predictor variable(s) and the SF, the
predicted variable. We find t=
hat
these ratios can help in auditing, detecting and deterring fraud, as well as
help develop rules for the ES software for Analytical Review (ARES). We app=
ly
it to one company and one industry, but design it to be more generic, so we=
can
expand it to other companies and industries. This study simulates fraud (SF) in=
the
financial statements. This Simulated Fraud (SF) combines Variable Fraud (VF)
and Fixed Fraud (FF) into a Mixed Fraud (MF). We regress, correlate, &
decompose such a MF into its VF & FF components. We compare accounts,
ratios, periods & companies. We regress a Focus company with Simulated
Frauds (SF) on its Peer Review Group (PRG) average (assumed to be without t=
he
same MF). We apply RAT to Expert System (ES) software and then test the
hypothesis that RAT can decompose MF & help in flagging the MF's riskie=
st
financial ratios. Eventually, the riskiest ratios will be part of the riski=
est
accounts.
Applying RAT, we decompose the MF into its =
VF,
& FF components. To decompose the MF, we regress it on sales. We define=
the
FF as regression’s intercept, and the VF as the regression’s sl=
ope.
Using the Least Square Regression we construct the MF function. We forecast
sales as a function of serial date, testing this date as a surrogate for sa=
les,
whenever it is a good predictor of sales. We construct a statistically
significant fraud model. This study calculates risk for Analytical Review
Diagnostics Controls for Automated Personal Computer (PC) Expert Systems (A=
RES)
based RAT. We integrate RAT, and the decomposition of this MF into the desi=
gn
of ARES, demonstrating some rules, screen, and reports that will result from
applying these ideas. By scanning financial statements and their ratios, and
regressing them against the peers averages the ARES could fire some rules.
These rules will quantify the risk of the likelihood of over or under stati=
ng
balances of accounts, helping auditors plan their audit. We could use such =
risk
measures to allocate audit time to accounts.
Likewise, such fraud coefficients help the =
ARES
prioritize its output and rate its opinions (related to problems in differe=
nt
accounts: phony sales ore receivable) from the most to the least risky
accounts. Thus, the ARES could minimize over loading the user with reports =
that
they cannot process. This way the users can limit the ARES to report only on
the top most risky accounts and/or ratios, or pages of reports. This will
relieve the users from evaluating hundreds of accounts and ratios. This is
especially true when the users are not knowledgeable, do not have the resou=
rces
to conduct such an evaluation, and cannot set apriority the materiality lev=
el.
This could facilitate the deployment of ARES for inexperienced managers and
auditors. Next, we would like to define the Simulated Fraud (SF) and some of
its assumptions.
Activity
Based Costing (ABC) and Cost & Management Accounting Theory (CMAT)
Applying
Activity Based Costing (ABC) theories to fraud, we develop the Activity Bas=
ed
Fraud (ABF) theory. ABF Treats
fraud as a cost that we want to trace back to the activities that generate =
it,
much like any cost. Except th=
at for
fraud, that unlike traditional costs, which we simply want to minimize, we =
want
to completely eradicate and eliminate, approximating a zero level fraud. To
calculate the cost effectiveness of fraud controls, we have to estimate it
(since criminals never report frauds voluntarily -- unlike other costs),
decompose it, leading to the activity sources. For that purpose we deploy t=
he
Reversed Accounting Theory (RAT) to trace the fraud back to its sources as a
step to eliminate such fraud. Our unique contribution is going in
‘reverse’ to traditional auditing. Instead of going from the
individual transactions to the financial statements, we go in reverse, from=
the
financial statements trying to reconstruct fraudulent entries. Hence, we
develop the concept of RAT. Much like reversed engineering, RAT goes in rev=
erse
to the common sequence of accounting work. We test the hypothesis that RAT =
can
construct a statistically significant fraud function.
We deploy the Cost & Management Accounting Theory (CMAT), and define VF, F= F, & MF, as well as develop software integrity controls. CMAT suggests that the slope of Mixed Cost (MC) regressed on the sales denotes Variable Cost R= ate (VCR). Likewise, the intercept of MC regressed on sales estimates the Fixed Cost (FC). Similarly, our Dollar SF regressed on the sales has a slope and = an intercept. This slope shows the Variable Fraud Rate (VFR), which is analogo= us to the VCR. Likewise, the intercept (of the SF on sales) shows the FF, analogous to the FC. RAT contends that whenever we are dealing with unknown fraud. Decomposing this fraud (into its fixed and variable components) will= be helpful in classifying its behavior and eventually pinpointing its sources.=
FRAUD
SIMULATION (FS) DEFINITION
fBARCLAYS
BANK PLC NET SALES - FRAUD
SIMULATION DEFINITION &nbs=
p; &=
nbsp; &nbs=
p; &=
nbsp; &nbs=
p; &=
nbsp; &nbs=
p; &=
nbsp; &=
nbsp; &nbs=
p; &=
nbsp; &nbs=
p; &=
nbsp; (---
See Screen2/ Appendix 1 ---)<=
/p>
Variable Fraud (VF), Fixe=
d Fraud
(FF), & Mixed Fraud (MF) Definition &n=
bsp; =
; &n=
bsp; =
; &n=
bsp; =
; &n=
bsp; =
; <=
/span> The 10%
{A31} Simulated Fraud (SF) of the NET SALES
{B33} primary (1st) account is balanced against the SF of the secondary (2n=
d)
balancing RECEIVABLES {B34} accoun=
t. The
FF remains fixed starting from the 1st period, 12/31/85
{L36}, to the last period, 12/31/94
{C36}, at $1,318,800 {C38} amount.=
We
have calculated dollar fraud amount as the
higher of 10% {A31} of the 1st pri=
mary
SF account or 1% of Net Sales. Add the FF to the VF and you get the MF for =
the
last year $2,637,600 {C39}. Add th=
is MF
to the initial balance of the 1st account, real, rNET
SALES {A55} balance $3,188,000
{C55}, for the last period. You get the phony, fraudulent, balance of 1st
account, fNET SALES {A54}, $15,825,600 {C54}. Likewise, for the 2nd
balancing account, start with the real balance, rRECEIVABLES
{A66}, $115,356,000 {C66}. Then, c=
ombine
the SF, $2,637,600 {C67}, and you =
get
the phony balance of the fRECEIVABLES<=
/u>
{A65} balancing that SF at $117,993,600
{c65}. Our question is which ratio is this fraud’s top predictor? fNET
SALES
Fraud, Why Would Criminals Create It & What Predicts It?<=
/p>
The objective and the res=
ult of
such a fraud may be the INFLATED NET SALES & RECEIVABLE, TO ENLARGE GROSS PROFIT &
STOCK MARKET VALUES {B31}. However, this may not explain why a
criminal would do it in a certain way. We could understand the possible pur=
pose
of such a Fraud by looking at the financial results. A fBARCLAYS BANK PLC {A30} executive working =
in the
Commercial =
Banks,
nec {C30} industry (COB
{B30}) may want to raise net income (reduce loss), credit ratings, commissi=
ons,
or promote his or her reputat=
ion.
For example, a fraud that overstates the Sales balance, will in turn overst=
ate
the Net Income. Overstating Net Income will make the company appear to be m=
ore
profitable. Such an overstatement of the Sales and profitability of the com=
pany
may benefit employees who own stock options, and/or commission employees. An
overstatement of Sales may easily translate to a pay raise, for certain
employees. Misclassifying Long Term (LT) assets as current assets, can impr=
ove
the credit rating of a company. However, such a fraud may not affect the in=
come
statement at all. We expect to find a ratio that can forecast such fraud.
10% fNET=
SALES
fraud: Account And Amount? (---See Screen 2/Appendix1---) A fraud perpetrat=
or
may prefer to misstate an account with a balance greater than zero. It is
possible, that overstating an account with a positive balance may appear a =
bit
easier to conceal, compared to an account with a zero balance. In the case =
of fBARCLAYS BANK PLC, and the Commercial Banks, nec industry (COB), the <=
u>rNET SALES (r=3DReal) account, is more like=
ly to
have a positive balance, compared to other companies & industries. An
account such as Cost of Goods will have a positive (debit) balance. Unlike =
some
other accounts, such as Investment Gains/Losses=
,
this is equally likely to have either balance, positive (debit), negative
(credit). Likewise, it is most likely to have a zero balance in a non banki=
ng
or finance environment. Once we have picked up an account, the next questio=
n is
concerning the amount. The Criminal (the perpetrator of the fraud) may not =
know
in advance what the normal balance of the account should be at all times.
Therefore, a fraud that is a percentage of the real balance may appear safer
(easier to conceal, automate, and harder to detect) choice. Likewise, small
percentages such as 10%, as used in the present case, may appear to =
be
safer than larger percentages. If the Criminals got away with the fraud for=
one
period, their greed may lead them to try it again and again. Therefore, a
multi-period fraud is likely. The Real Peer Co. rNET SALES (Source
Account) & Its Phony (f) Partner (--- See Screen1/Appendix1 ---) The common criminal would=
make a
fraudulent entry, where the Debit and the Credit do balance. Some accounting
software may prevent unbalanced entries. Such entries may also be too obvio=
us,
and too easy to detect. Therefore, to supplement the present fraudulent ent=
ry,
the Criminal should make a balancing entry in other accounts that normally
balance the source account. Such an account could be the (Focus Co.)
Reversed Accounting Theor=
y (RAT)
will help us pinpoint the source of the fraud a=
nd
estimate its damages. This should help us detect it, if we can possi=
bly
classify it correctly (variable versus fixed). For example, if the fraud
positively correlates with sales, or varies proportionately with the sales
account, that gives us a clue as to which account actually contains this VF.
Such a fraud most likely originates from sales itself, or one of its
derivatives, such as sales commissions. In contrast, we can more easily
LITERATURE REVIEW
Activity Base Cost / Fraud (ABC/F), Short and Long Runs, & Regression Analysis
The developments in activity-based costing (ABC) since the late 1980s have improved Accounting Information Systems. However, no one has ever applied ABC to Fraud, which is the focus of the current study. Applying ABC to Fraud produces Activity-Bas= ed Fraud (ABF). The literature covers other extensions of ABC to management, or Activity Based Management (ABM), but not to ABF. Hartnett and Lowry (1994) predict total cost for change in product mix. We are also trying to predict costs, except that we predict the cost of fraud damages as a function of fr= aud mix (VF, FF, & MF). Holmen (1995) suggests that ABC has primarily a long-run horizon. Therefore, we apply it to long-run fraud estimation probl= ems, frauds that continue for 1 year, and usually much longer. Macintosh (1994) suggests that in the “scientific&nbs= p; ABC” method the designer uses multivariate regression analysis.
Hartnett, Lowry, and Luth= er (1994) confirm that ABC supplies better approximations of long-run variable costs. This is one of the reasons that we are focusing strictly on long term frauds. Likewise, we suspect that ABF may also lead to similar results; it = may supply better results in long run, rather than short run analysis. Even if we detect no material frau= d, just by estimating the fraud we may reduce its damages by acting as a deterrent. There are, however, obstacles to using ABF, much like ABC in reporting. Such obstacles include applying standards rather than actual val= ues, subjectivity and verifiability in the choice of cost-drivers, and ABC or ABF integration into = the nominal ledger. We use the pre-fraud as surrogates of fraud free standards, normally based on competitors' financial statements. The problem is that selecting proper competitors is subjective and may vary greatly depending on the industry, economy, and other variables.
Pattison, Arendt, and Gav= an (1994) implement a modified ABC system. We have modified ABC so much that we have to rename it ABF. It is too different from ABC, to retain the same nam= e. Sheu and Wacker (1994) integrate time series forecasting, and ABC analysis. Similarly, we are doing the same thing. We also integrate time series into = ABF. Datar and Gupta (1994) suggest that an ABC generates more accurate product costs than other systems. For a similar reason, we hope that ABF will gener= ate more accurate fraud cost estimates than other systems. Groth and Kinney (19= 94) suggest activity-based costing (ABC) and cost driver analysis may reduce business risk, promoting value creation in a firm. Similarly, we contend th= at ABF and its cost drivers may help reduce fraud risk. In a counter sense, fr= aud cost management not implemented properly result in an intensified eradicati= on of value. These reasons attest to the importance of studying fraud costs in value creation. This may ultimately lead to a cost benefit analysis of fraud eradication investment.
ABF as A Control for Long= Term Fraud & As a Function of Sales
Mak and Roush (1994) argu= e that ABC controls costs. Likewise, we feel that ABF controls. Under one proposal, they formulate flexible budgets for each activity using the cost driver for that activity. We are using sales or some of the financial ratios as the surrogate cost drivers of the fraud in order to develop an estimation metho= d. In the future, we may try other cost drivers. Smith (1994) extends ABC to a management context and calls it: activity-based management (ABM). Similarly= we extend ABC to ABF. Our extension is a bit less radical in the sense that we can view fraud as a co= st category, while management is a much broader concept than cost alone. We all agree that most of the emphasis in the literature, thus far, has been on co= sts (activity based costing) and quality (total quality management). We also consider other areas and factors. A time-based focus has a number of positi= ve implications for the management accountant in designing improved management= information systems. This ensures that decision-making is linked to an appropriate time horizon by matching short-run and long-run costs with decisions. For these reasons we focus on the long-run in the current model, reserving the short-= run for future studies.
Literature of Fraud, Damages, Detection Exp=
ert
Systems (ES) & Disclosure
A major problem in fraud detection i= s the lack of education on the part of those who must detect it (Kerwin, 1995). S= ome early warning signs may be found by parsing financial statements (Donlan, 1994). We combine a Simulated Fraud (SF) educational approach with parsing financial statements to find the best fraud predictor as warning signs in o= ur ABF theory. Redirected Cash flows the= wrong receivables or the wrong disbursement location is a common early warning si= gn of a cash fraud (Marks; Arnette, 1994). Employees commit the overwhelming majority (90.8%). Although at a much lower rate, executive fraud is = only 26% (Campbell and Lindsay, 1994). Therefore, we our SF deals with an employ= ee fraud, such as a cash fraud. The Fina= ncial Fraud Detection and Disclosure Act, requires exception reporting when control systems fail, such as material financial frauds (Campbell and Linds= ay, 1994). Neural networks can help to find patterns and relationships, even obscure and nonlinear relationships (Stewart, 1994; Basch, 1994; Mayor, 199= 4). In our SF the relationships are fairly linear; therefore, we use linear regression. One way to combat managem= ent fraud involves Analytical Procedures (AP). Quantitative APs alone will not detect fraud; they simply signal the likelihood of a problem (Calderon, &am= p; Green, 1994). This is where our model comes in. Our model actually estimates the dollar damage of the fraud, after the AP signaled a high likelihood of a problem.
OVERVIEW & DEFINITION OF PR=
OBLEMS
The problem is that we do not know how to
eradicate and prevent fraud. In addition, we do not know even how to estima=
te
fraud and how to trace its sources, which may be a prerequisite to preventi=
on.
Estimating fraud and tracing it back to its initial transaction is the focu=
s of
this study. To find the initial fraudulent transaction we developed the
Reversed Accounting Theory (RAT). RAT states that, unlike ordinary accounti=
ng,
where we start from individual transactions and conclude with summarized
financial reports, the reversal may be more effective for fraud estimation =
and
definition. Thus, RAT starts from the summarized financial reports, which m=
ay
be conceal fraudulent transactions, concluding with estimating the value of=
the
fraud and defining the fraudulent transaction.
To define the culprit transaction, we use t=
he
PERCENTAGE Differenced Financial Ratios (%DFR), such as the Dollar Sales per
Employee or Quick Ratio, instead of the original accounts, such as Cash or
Inventory. There are some adv=
antages
to using %DFR instead of using the original accounts. There are fewer %DFRs
(dozens) compared to the original accounts (hundreds). The process of
eliminating the irrelevant suspects is easier. The %DFR are much more
standardized and uniform across times-series, companies and industries, than
the original accounts themselves. The relative (Percentage) unit of measure=
ment
is more comparable than the original unit of measurement (inter or intra
company or industry or currency comparisons of $ Sales per Employee to the
Quick Ratio)
Our accuracy of spotting and forecasting the
existence of frauds is “extremely good” (Fanning et. al., 1995,
Coates et. al., 1991). Likewise, we know what the red flags that indicate
perpetrated frauds are (see: Fraud Red Flags Questionnaire, Pincus, 1987). =
This
sounds sufficient to scare auditors about being held liable for not reporti=
ng
the fraud (AICPA, 1988). Yet, such indications are much too vague to enable
auditors, prosecutors, or managers to pinpoint the fraud. We still do not k=
now
how to estimate that fraud, what activity generates such frauds, and which
transaction initiates such fraud. We think that if we can figure out the
activity that generates the fraud, we should be closer to tracking the
fraudulent transaction, and eventually the culprits themselves. For that we
propose an Activity Based Fraud (ABF) theory.
We view fraud as a cost item. Fraud is cert=
ainly
not a revenue item, nor is it a liability, assets or an equity item. Therefore, it is most similar to c=
ost
items, even in the sense that we are undoubtedly trying to minimize fraud (=
at
least in theory). In fact, fraud may be the oldest cost, from biblical time=
s,
yet we do not estimate and account for it like any other cost. Fraud is
currently the leading cause and cost of auditing litigation (Palmrose, 1991=
; St
Pierre and Anderson, 1994). Auditors have changed the way they operate due =
to
the increase of litigation (O’Malley, 1993; Elgin, 1992; Fuerman, 199=
2).
Auditors charge escalating fees to fund such litigation risk.
Based on Activity Based Costing (ABC) theor=
ies,
every cost originates from some activity. Such activity is the cost driver,
which are among the best cost predictors. Therefore, if we can treat fraud =
like
any other cost, we may apply ABC to fraud. Like any other cost, certain
activities produce fraud. Therefore, it only follows that like any other co=
st,
every fraud may have its own drivers, which are its best predictors. We try=
to
develop a method of identifying these drivers for specific types of fraud,
using ABC methods.
Unlike any other costs that are legal, frau=
d is
illegal. Therefore, we cannot simply use traditional transaction-to-financi=
al
reports accounting. Thus, we deploy the RAT approach and combine it with AB=
C.
From this, we create the Activity Base Fraud (ABF) theory. ABF states that =
if
we can find the Best Predictor %DFR (BP%DFR), we will be further along the =
way
of estimating the fraud, its drivers, and eventually it original culprits.
Furthermore, by just deploying ABF, it may act as a deterrent to fraud. Jus=
t as
we are using standards to cost the unknown overheads, we may eventually also
use standards to cost the unknown frauds.&=
nbsp;
We have some red flags that fire up wheneve=
r the
likelihood of multi-year fraud rises. Others have clearly defined such flag=
s (discussed
above), so that not only an auditor could use it, but also a computer base
Expert System (ES) could use it, while it is emulating an auditor. Our curr=
ent
objective is one step further, to estimate the amount of the fraud, and its
original transaction or activity that created that transaction. For example,
such an amount could be a billion dollars of overstated marketable securiti=
es
over several years. Such was the case of unrecorded investment losses from
futures trading at Diawa bank. Another scenario is phony sales debited to
accounts receivable. A fraud like this was committed in the Equity Funding =
case
where sales were in the names of the deceased.
For this purpose
we have simulated multiple frauds (SF=
) on
the financial statements of this company. The simulated Frauds (SF) include
increasing net income over the last few years. Thus for example, we could h=
ave
overstated sales, and understated all the expenses by 10% of their values. =
At
the same time we have also entered a balancing entry, so that the debit and=
the
credit remain balanced, and do not obviate the frauds (see Appendix for
details).
We have then benchmarked the SF company aga=
inst
its own financial ratios prior to the SF. We use that to emulate how a real
fraud ridden company may compare to its peers, which are not involved in the
same type of frauds. This way we hope to identify a methodology of spotting=
red
flags and patterns to detect and deter fraud.
COMPANY
AND INDUSTRY LITERATURE REVIEW
OVERVIEW OF BANKING INDUSTRY
At the nation's 11,970 FDIC-insured institutions, total assets grew $84 billion from September 30 = to December 31, compared to growth of $90 billion during the fourth quarter of 1994. Fourth-quarter asset growth was funded to a greater extent by deposit= s in 1995 than in 1994. Total deposits increased $96 billion during the fourth quarter of 1995, with most of this growth occurring in domestic deposits at commercial banks, which were up $91 billion. In 1994's fourth quarter, by comparison, total deposits increased $70 billion, including just $42 billio= n in domestic deposit growth (Helfer, 1995).
The reserve ratio of the = Bank Insurance Fund (BIF) was 1.30 percent of insured deposits on December 31, d= own nominally from 1.31 percent on September 30 but still above the statutory minimum of 1.25 percent. The reserve ratio of the Savings Association Insur= ance Fund (SAIF) rose from 0.43 percent to 0.47 percent during the fourth quarter but remains far below the target level of 1.25 percent. As a result of the BIF's full capitalization, the FDIC was able to reduce BIF assessment rates twice in the latter half of 1995. The average BIF assessment rate fell from 23.3 cents per $100 of assessable deposits to just 0.4 cents (effective Jan= uary 1, 1996), improving the attractiveness of BIF-assessable deposits relative = to other funding alternatives. Because SAIF assessment rates cannot be lowered significantly until the fund is fully capitalized, the average assessment r= ate for SAIF members remains at 23.7 cents per $100. (1) This premium disparity between the BIF and the SAIF may partially explain deposit growth patterns = in 1995. Deposits assessable by the BIF grew $110 billion (4.6 percent) during 1995, with $83 billion of the increase coming in the fourth quarter. Deposi= ts assessable by the SAIF grew $20 billion (2.8 percent) in 1995 but decreased= , by less than $1 billion, in the fourth quarter. (2) Other factors, such as loan demand, also may affect deposit growth patterns (Helfer, 1995).
BARCLAYS BANK COMP= ANY PROFILE
Group profit before tax i= mproved by 1,198m as a result of a decline in provisions for bad and doubtful debts= . Of this reduction, 856m occurred in the United Kingdom, where new gross specif= ic provisions were 791m, and 309m in the United States, where new gross specif= ic provisions were 97m. Although falling significantly, non-performing lending remained at a high level, particularly in regard to the UK property, construction, hotel and leisure sectors.&n= bsp; As a consequence of work that is being undertaken to improve the ass= essment of credit and credit losses throughout the business, the general allowance = to cover unidentified bad debts has been increased by 74m to 850m (Barclays (b= ), 1994).
Profit before tax showed a significant improvement over the two previous years as budget debt provisio= ns fell, helped by releases in the United Kingdom and the rest of Europe. The reduction in provisions was achieved despite one large corporate provision = of 65m in Europe and further provisions in Canada (Barclays (b), 1994).
FUTURE OF BARCLAYS BAN= K COMPANY
The Trust Company profit = for the year was adversely affected by an ongoing re-organization process, the resu= lt of which will be to improve efficiency and customer service in the future. = Most of the Group's trading activities are customer oriented. In anticipation of future customer demand, the Group maintains access to market liquidity by q= uoting bid and offer prices with other market makers and carries an inventory of capital market instruments including a variety of derivative and non-deriva= tive (or 'cash') financial instruments. Positions are also taken in the interest rate, foreign exchange, debt, equity and commodity markets based on expectations of future market conditions. These activities constitute the proprietary trading business of the Group. Given the relationships between instruments and markets, trading strategies depend on both market-making and proprietary positions and are managed in concern in order to maximize tradi= ng related revenue. Trading positions and any offsetting hedges are establishe= d as appropriate to accommodate customer or Group requirements (Barclays (a), 19= 94).
FUTURE OF BANKING = INDUSTRY
Five Paces, Inc. has deve= loped the next step in banking technology: Virtual Bank Manager (VBM). VBM is a software solution that allows financial institutions to conduct secure on-l= ine transactions over the Internet. It is the first module within Virtual Finan= cial Manager. We're developing additional modules that can be incorporated within Virtual Financial Manager or, like Virtual Bank Manager, can be used as a stand-alone solution. VBM w= ill allow anyone anywhere with a PC easy access to full-service, on-line Intern= et banking. (Five).
METHODS AND PROCEDURES, DATA ANALYSIS &
INTERPRETATION
DATA DEFINITION, INPUT, &= amp; OUTPUT OF REGRESSION ANALYSIS THEORIES
Differenced Financial Rat= ios (DFR) Measure Fraud Impact & Identify Fraud Insensitive Ratios
We down load the company = Profile, Annual Report, & Financial ratios from the Disclosure database and then upload them into our Fraud Simulator spreadsheet Software Program (The Program). The Program recalculates all these Financial Ratios verifying the= ir values and extending their decimal point. Then the program applies the Simulated Fraud (SF) to the recalculated and verified Annual Report, produc= ing a second set of phony fraudulent fAnnual Reports. The Program inputs the fAnnual Reports to the financial ratio calculation and produces a second se= t of phony fraudulent financial ratios. The difference between the 1st real fina= ncial ratios and the second fraudulent financial ratio produces the Difference Financial Ratios (DFR). Such DFR measures the effects of a fraud on that particular ratio. If the DFR equals zero, then we can conclude that a particular fraud does not affect it. Likewise, if the DFR of another ratio = also equals zero, then we can conclude that like the 1st ratio, the same fraud d= oes not affect the second ratio. = Thus, both ratios are insensitive to this fraud. The DFR is only comparable when = it is zero. When it is different than zero, we cannot use it effectively for comparisons.
DFR Inability To Compare = Fraud Impact & Identify A Given Fraud Most Sensitive Ratios
These DFR denominate their original values and therefore are hardly comparable. For example, the Net S= ales Per Employee (NSPE) DFR can describe the fraudulent increase in sales per employee due to the fraud. Likewise, the DFR Quick Ratio (QR) describes the change in liquidity due to the fraud. However, the absolute value of the NS= PE DFR will always exceed its QR counterpart, since NSPE it measures Sales per employee. Thus, we cannot tell which one is more sensitive to this particul= ar fraud, just by comparing their DFR values. Since one of our objectives is to find the ratio that is most sensitive to this kind of fraud, we have to dev= elop a more relative measure. Such a measure will compare the sensitivity of different ratios to different fraud simply based on its magnitude.
Relative %DFR Comparison = Fraud Impact & Identify Most Sensitive Ratios
We define the relative %D= FR as the DFR divided by the real Financial Ratio. This way we can decide which r= atio is more sensitive to a given fraud, even though these ratios denominate completely different units of measurement. While the NSPE DFR and the QR DF= R are clearly not comparable, their counter parts, the NSPE %DFR and the QR %DFR = are clearly comparable. Comparing their magnitudes, describe their sensitivity = to the fraud. Typically, the ratio with the highest %DFR will also be the most sensitive to that particular type of fraud. This way we can construct a Financial Ratio Fraud Sensitivity Theory (FRFST) and use the %DFR to test o= ur theory.
Financial Ratio Fraud Sen= sitivity Theory (FRFST) Tested By DFR, %DFR, Or Both
We can hypothesize that a= fraud that overstates (credits) Credit Sales, balancing it by overstating (debiti= ng) fixed assets, should affect the NSPE ratio more than the QR. To test such theory, we can simply use the DFR value of both ratios. Since this fraud wi= ll not affect any current assets or any current liabilities, its QR DFR must be zero, while its NSPE DFR will be larger than zero, and therefore more sensitive. But, what if The Fraud balances the Credit Sales by overstating (debiting) Accounts Receivable, which is a much more likely fraud. This way= one of the DFRs will no longer remain zero, and will no longer be as comparable= as the former example. In this case, we can call %DFR to the rescue, using it = to test our hypothesis and further construct while validating our FRFST. Can we therefore stop using DFR and simply replace it by %DFR? Not necessarily so, some time DFR may be necessary, since we cannot calculate %DFR.
Fraud Damage Cost Estimat= ing Software Program (FDCESP) & DFR/%DFR 2 Stages
Due to the difficulty that computers have in calculating divisions by zero, it is sometimes necessary = to initially use a measure like DFR. In developing a Fraud Damage Cost Estimat= ing Software Program (FDCESP) we have to be considering such issues. Since we w= ant to extended current ES technology with such a FDCESP system, we have to consider machine difficulties. Furthermore, since we typically want to mini= mize computer resources and maximize efficiency, we would use DFR as well as %DF= R. We have to calculate DFR before we can calculate %DFR. Therefore, we will a= lso use it in our FDCESP development as a 2 stage process. The Program will fir= st calculate and use DFR, if it can choose the most sensitive ratio. If it can decide at this stage, it= can then store its decisions in a Case Based Reasoning Knowledge Base (CBEKB). = The program needs to go no further. It does not need to calculate %DFR. This ma= y be a case where the DFR will be zero, anyway. Therefore, we do not need to calculate %DFR for a decision. Furthermore, the attempt to calculate it may result in a Division By Zero error, creating all kinds of problems.
Case Based Reasoning Know= ledge Base (CBEKB), Expert System (ES) & FDCESP
In this context, a Case B= ased Reasoning Knowledge Base (CBEKB) is a data base of fraud cases and its rela= ted decisions. Such CBEKB can be a part of an Expert System (ES) that issues a Fraud Detection, Damage Estimation, Investigation, & Prevention opinion= s. Such opinions will rely on a Cost Benefit Analysis comparing the benefits of fraud damage reduction to its costs. This is where Fraud Damage Cost Estima= ting Software Program (FDCESP) integration fits. For a system like this to work, identifying a few best predictors of fraud value could be very useful. These variables will be the Best Predictors PERCENTAGE % Differenced Financial Ra= tio (BP%DFR).
Best Predictors Percentage Differenced Financial Ratio (BP%DFR) Flags Fraud Patterns
Best Predictors Percentage Differenced Financial Ratio (BP%DFR) variables will estimate the dollar (or other currency) value of damages. Such analysis should be done when existin= g ES predict that the financial statements do contain fraud, and the main questi= on is which kind of fraud is it? Is it a Credit Sales and Fixed Assets fraud? = Is it a Credit Sales and Accounts Receivable fraud? This is where various BP%D= FR can red flag the various types of frauds that may be present. As we can reasonably expect, different ratios will become BP%DFR for different types = of fraud. For example, if the fraud pattern appears to have fixed values over time, then it may be a Fixed Fraud (FF). A FF is more likely to origin in f= ixed assets accounts, rather than in a Sales Commission account. In contrast to = FF, a Variable Fraud (VF) is more likely to originate in a Sales Commission acc= ount or may be in a Direct Labor Cost type of account. Likewise, a Mixed Fraud (= MF) may be split over both types of accounts, etc. Furthermore, this BP%DFR will estimate the total dollar amount of the fraud telling us how many resources will it justify, and most importantly which accounts are the culprits.
Regressing Fraud Values O= n Best Predictors % Differenced Financial Ratio (BP%DFR)
Regressing Fraud Values O= n Best Predictors Percentage % Differenced Financial Ratio (BP%DFR) will help buil= d a model to estimating Fraud Values. To set up the data for such a regression analysis, we split the data into 2 parts. The first part constitutes the pre-fraud condition. Naturally, for these observations, the independent predictor variables, the BP%DFR, as well as the dependent predicted variabl= e, the Fraud Values, are both zeros. In an experimental simulation, that is certainly true. This is saying that if no fraud what so ever exists, then t= he pre-fraud and the post-fraud ratios will be the same. Thus, the BP%DFR, as = well as all other differenced ratios, must equal zero. In contrast, in the real world, if no fraud what so ever exists, then the Peer Review Company, or Companies and Industry Averages (like pre-fraud, free fraud) and the Focus Company (like the post-fraud, fraud ridden) ratios will not necessarily be = the same. However, they may approximate identity, if the peers are almost ident= ical to the focus company. Although, identity between a focus company and its pe= ers is very unrealistic condition, there are ways to approximate such condition= , by statistical and mathematical means. Such methods are beyond the scope of the present study, however, we will discuss them briefly later, and in future studies. The second part of these observations will typically be different = than zero. These 2 parts will make up 2 parallel time series. One time series without fraud, and the second with fraud. Using these 2 sets of time series= in parallel, makes this analysis a cross sectional and time series regression analysis.
DATA INPUT FOR THE REGRES= SION ANALYSIS: ITS EMPIRICAL ASPECTS
&n=
bsp; =
; &n=
bsp; =
; &n=
bsp; =
; &n=
bsp; =
; <=
/span> (--- See Appendix 1 in
REGRESS.XLS)
&nb=
sp; =
&nb=
sp; =
&nb=
sp; =
For
the top predictor, X1 BP%DFR, X1=3DNET
SALES/TOTAL ASSETS, we com=
pute
the 2nd part of the positive fraud observations, as follows. Let’s pi=
ck
the 12/31/86 period, since this period maximizes the BP=
%DFR
value. Following are steps that calculate some of the predictor variables'
values, such as X1, X2, etc. =
We
start with the first variable, X1. The fraudulent, phony, fNET SALES/TOTAL ASSETS {a74}, has a value of 0.13049=
{M74}. In contrast, the real,
You
calculate the MF Cost Function (MFCF) by regressing VF, FF, & MF on Net
Sales. Start with VF. The VF regression has an R-Square of 1 {b37}, an Intercept of $ - {=
m37},
and a Slope of 10.000000%
Theoretically,
for a Pure VF definition, the Intercept should be a perfect zero, but that =
may
change due to rounding errors. A Gross VF definition may cause an error, if=
we
define the VF as a percentage of a primary account (not perfectly correlated
with Net Sales). Such an account may be Long Term Assets. The FF should be =
the
exact complement (mirror image) of the VF, a slope of zero, and an Intercept
that is positive. At the same time, the VF should have the opposite values,
zero intercept and a positive Slope. Their sum, MF, should have both positi=
ve
Intercept and Slope. Likewise, we made MFCF by combining the separately
regressed VF & FF. These 2 functions (VF & FF) should have the same=
2
regression parameters (intercept and slope). Much like the single MFCS regressi=
on, it
should also have an intercept and slope. These redundant calculation contro=
ls
demonstrate how to promote software integrity. Classify and decomposing fra=
ud
into its parts will help detect, measure, and ultimately minimize fraud dam=
ages.
A pure version of the definition may be more effective than the Gross versi=
on,
but any one is better than none at all.
The R-Square shows the
statistical quality of this model, which we will discuss later. Such MFCF c=
an
estimate fraud prevention benefits. It traps error by comparing actual and
expected results. Thus, the growth rate & variance of FF should be zero=
. If
it indeed is zero, then we may have avoided this error. In this case, these error values are equal to: 0.00% {c41}. Ev=
en if it
is not zero, it may be due to an immaterial error such as a rounding error,=
or
an error in the Disclosure data base, that is beyond the scope of this stud=
y. (---
See Screen 4 ---)
Expert Systems (ES) exemp= lify regressions' rules. The ES can regress the estimated fraud values (estimate= d by our model) on the NET SALES account of the Focus company. If the regression= is significant, then the ES may reach some tentative conclusions, about the possibility of fraud. This can be true, if it turns out to be a case of a P= ure definition of VF, FF, & MF. The ES further reinforces its conclusion th= at fraud may exist, estimating its cost function structure, dollar estimates, = and probabilities. Additional tables will reveal some more analysis to further reinforce such suspicions, and eventually come up with an explanation to the fraud. ES show that explanations to the Variable Fraud (VF) can include: “10% INFLATED NET SALES & RECEIVABL= E, TO ENLARGE GROSS PROFIT & STOCK MARKET VALUES.” Yet, we n= eed more analysis. Our methods may help ES calculate risks and correctly interp= ret them. Are the ratios correct? Is it Fraud or is it not fraud? That is the question. This ES tests itself, using= a value and sign test as an Internal Controls Of Integrity & Reliability. (--- See Screen 5 ---)
REGRESSION ANALYSIS & ANALYSIS OF VARIANCE (ANOVA)
This Regression Analysis = (RA) is a Linear Regression Analysis Of Focus Company Simulated Fraud On the Best Predictors Percentage % Differenced Financial Ratio (BP%DFR). This RA inclu= des an ANOVA, residual analysis, and Plots. The SUMMARY OUTPUT, part 1 of the R= A, contains the Regression Statistics including 5 of the most critical results= of the analysis. This RA contains 4 parts, 3 sub-analyses and 1 set of plots. First, is the SUMMARY OUTPUT and the Regression Statistics sub-table. Secon= d is the ANOVA (Analysis Of Variance) sub-table. Third is the Residual & Probability Output sub-table. Fourth is the set of plots including; a Resid= ual Plot, a Normal Probability Plot, and a Line Fit Plot. Next we will discuss these tables.
REGRESSION ANALYSIS: SUMM= ARY OUTPUT
&n= bsp; = ; &n= bsp; = ; &n= bsp; = ; &n= bsp; = ; <= /span>(--- See Table 1 in REGRESS.XLS)
[1]SUMMARY OUTPUT FOR BARCLAYS BANK PLC (THE = FOCUS COMPANY)
The Regression Statistics sub-analysis contains the 3 parts. The 1st part is the Multiple R, 0.999845, [Table 1] {B6}. The 2nd part is t= he lower R-Square, R**2, value of 0.99969{B7= }. The 3rd part, following the R-Square is the even lower ARS (Adjusted R-Square) = of 0.999632 {B8}. Multiple R is the correlation between the predicted and the actual values of the focus company BARCLAYS BANK PLC, the coefficient of correlation. These correlate the actual and the forecasted values of this Simulated Fraud (SM).
Expert Systems (ES) Form = Patterns Of Rules For R-Square’s Upper & Lower Bounds
&n= bsp; = ; &n= bsp; = ; &n= bsp; = ; &n= bsp; = ; &n= bsp; = ; (--- See Screen7/8/Appendix1 ---)
This regression analysis =
helps
Expert Systems (ES) form patterns of rules for fraud classification. The ES
will apply such rules to a Suspicious Company, to classify its fraud. We wi=
ll
form the ES rule in a step by step fashion as we discuss the output.
For the Regression of the Differenced Account On The Differenced Financial Ratios:
If the Industry is equal = to This Industry and the Suspicious Company R-Square is Equal To
Less than the upper bound= of R**2: 0.99969{B7}, And
Greater Than the Lower bo= und of ARS: 0.999632 {B8}, And è
In practice this will be = one large rule, for simplicity we break it into its components. Arrow heads indicate è that the rule continues later. A Knowledge Base System (KBS) will store these rules.
Multiple R (R) -- Internal Software Control, The Sum Of Square (SS) & Beta Values
This correlation ranges f= rom -1 to +1. If the Multiple R is equal to 1, then a change in the value of X, the independent predictor variable, the BP%DFR predicts the dependent predicted variable, the focus company, BARCLAYS<= /u> BANK PLC BP%DFR, perfectly well. Otherwise, Multiple R is different from 1,= and the prediction is less than perfect. The present Multiple R of 0.999845 [Table 1] {B6} tells us that there= is relatively strong = linear relationship. The independent variables (the BP%DFR of The Focus Company &a= mp; the COB industry) estimate the Simulated Fraud (SF) very well. This means that if the BP%DFR increases, then the Simulated Fraud (SM) is also v= ery likely to increase as well, in this sample of data. This value of R tells us two things. If this is statistically significant, it may apply to other samples.
R-Square: Variance In The Simulated Fraud (SM) Explained By The Financial Ratios
Unfortunately, we cannot interpret R in more precise terms, since it is not equal to 1. The R-Square, the Coefficient of Determination, may help us determine the meaning of the relationship of the population more precisely. Expert Systems (ES) would typically apply such relationship to the population of cases.
The Multiple R (R),
The Adjusted R Square: A = Function Of The R And The Number Of Observations
The Adjusted R-Square (AR= S), 0.999632, [Table 1] {B8} is lower than both= the Multiple R and the R-Square. We adjust the ARS for the degrees of freedom (= df). The number of Observations, 20 {B1= 0}, determines the df. This ARS tempers the strength of the R, which may be unrealistically high, especially if we have a small number of observations.= To avoid overstating the strength of R-Square, we calculate the Adjusted R-Squ= are. This difference between the R-Square (RS) and ARS will shrink as we increase the number of observations. We split these Observations into 2 parts. The 1= st half (10 Observations) includes Zero values of both the Predictor Financial Ratios, BP%DFR, and the Predicted Simulated Frauds. These zero values repre= sent no difference in financial ratios in the absence of fraud. Thus, the differ= ence between the Fraudulent and the Real Financial Ratios, as well as the Simula= ted Fraud itself is all equal to zero. The second half shows the differences in= the Financial Ratios that explain the Simulated Fraud. In contrast to the origin values of zero, these later 10 observations will usually differ from zero. Otherwise, we may not have simulated a fraud; we know this by definition and design.
ANOVA: ANALYSIS OF VARIAN= CE: Testing The Utility of the Model &nbs= p; (--- See Table 1 ---)
The difference between the R-Square (RS) and ARS will shrink as we increase the number of observations. If such differe= nces exist, then, the next sub table ANOVA portion will explain them. Specifical= ly, the SSE (Sum of Squared [SS, column] Error or Residual [row]), that is much greater than zero, explain some such differences. In our case, the value of= SS Residual is 1.1E+10 [Table 1] {C15= }, which is very small and very close to zero, explaining the proximity of RS & ARS. If RS & ARS are much further apart, the regression model may= not be as reliable (compared to close RS & ARS).
The ANOVA helps us evalua= te the utility of the entire model. It tests individual coefficients allowing us to conclude that a linear relationship between the independent and the depende= nt variables exists. The ANOVA decomposes&nbs= p; the variability of the dependent variable, The Emulate Fraud, measur= e by SS Total, 3.55E+13 [Table 1]{C16} = into its components, the Regression SS, 3.55E+13 {C14}, and the Residual part, 1.1E+10<= /u> {C15}. If the SSR (Regression) is large relatively to the SSE (Residual), t= he RS is high -- signifying a good fit and a good model, as is our present cas= e. The SS column enables us to calculate the Mean Square (MS) column. Their ra= tio, F equals (MSR/k)-(MSE/ [n-k-1]) column (Where n=3Dnumber of observations, <= u>20, and k=3Dnumber of independent variables= , 3 {B14}). The Degrees Of Freedom (df) for the Regression is the observations minus the independent variables. A value of F, 17225.41 [Table 1] {E14}, shows the significance of the model. This F value is statistically significant at the .05 level. That means that the BP%DFR, Xs independent variables, explain the variation in Y dependent variable, the Simulated Fraud Dollar amount= for BARCLAYS BANK PLC. Therefore, the model is = useful for estimating the cost for this kind of fraud damages (See Glossary For Definitions).
ANOVA PREDICTORS-TEST OF = THE INTERCEPT CONSTANT AND THE BETA SLOPE(S)
The test of the Intercept=
and/or
individual Beta coefficients allows us to decide about the linearity
assumption. We decide whether a linear relationship exists between the
Independent Predictor Financial Ratios and their Predicted Dependent Variab=
le,
the Simulated Fraud in the focus company, BAR=
CLAYS BANK
PLC. Consequently, we perform the t-test for the Intercept constant,
Intercept Coefficient: Ke= eping (Unable To Reject) The Null Hypothesis That The Intercept Is Zero
To evaluate the statistic=
al
significance of the Intercept Coefficient, -1=
3.1994
{B19}, we compare it to its Standard Error. The Standard Err column,
Expert Systems (ES) Form = Patterns Of Rules For ANOVA & Intercept’s Upper & Lower Bounds
Using the ANOVA bounds of= the Intercept, the ES can continue forming the rules as follows:
èIf The ANOVA is statistically significant = at the 5% level, & The Suspicious Company’s Intercept is:
Less T= han Intercept Coefficient Upper 95% Bound Of: 17558.35, And
Greate= r Than Intercept Coefficient Lower 95% Bound Of:-17584.7, And =3D=3D>
Expert Systems (ES) Form = Patterns Of Rules For Predictor (Independent) Variables: The
X1-3 Coefficients Or Slop= es, & Their 95% Upper And Lower Confidence Interval (CI) Bounds
We will plug the X1-3 Coefficients Or Slopes Of The Variables in the final regression equation. To complete the regression equation we need also the residual. Therefore, we w= ill do that in the after the residual output analysis. In the mean time we can complete the Fraud Pattern ES rule, using the 95% Upper And Lower Bounds of= this variable together with the R-Square and the Intercept, as follows:
&n= bsp; = ;
=3D> the 1st Best Predictor is: X1=3DNET SALES/TOTAL ASSETS {A20}, And
its Coef= ficient (Slope) value is Greater Than 95% lower Bound of: -1.5E+07, And
its Coef= ficient (Slope) value is Less Than 95% lower Bound of:-486124517558.35, And
The 2nd Best P= redictor is: X2=3DSG & A/SALES {A21}= 230;…. (same as above but for X2 Bounds), And
The 3rd Best
At this point, the ES wil= l also fire an explanation as to support its decision (See Screen Shots for a demo= ).
&n= bsp; = ; &n= bsp; = ; &n= bsp;
To complete the calculati= on of the estimated fraud, we need to include the Residual value, in addition to = the values of the financial ratios. Thus, we will replace the X symbols by their values, and add the residual to the sum, that should produce the Simulated Fraud in a given year. In contrast to the insignificant intercept, the slop= es are all significant. Thus, we reject the null hypothesis that the slopes are all equal to zero. We can plug our parameters into the model after we figure out the residual value.
RESIDUAL & PROBABILITY OUTPUTS: The Plots Of Normality, Residuals, & Variances
The Least Square method o= f the regression analysis requires meeting the following conditions. (1) The error variable in the residuals is normally distributed. (2) This error has a mea= n of zero. (3) It has a fixed variance and (4) the values of the residuals are independent. To diagnose most departures from these assumptions, we view the RESIDUAL OUTPUT & PROBABILITY OUTPUT as well as the Residual, Line Fit, & Normal Probability Plots. Using the ANOVA for the intercept (a) and slopes (X1, 2, 3) we include the residual value, and we can construct a complete linear regression model, calculating Y.
The following is an examp= le of a 20 Observation series BP%DFR model including the residual (which contributes the last term of this regression equation, other terms come from the ANOVA)= :
$ 3032408 The SM (Y, predicted value) {B43} =3D -13.1994 {B19}(a=3D Intercept constant) + <= u>-1E+07 {B20} (b1, Slope) * (X1=3D 15.44777% {Appendix1 B15}: X1=3DNET SALES/TOTAL ASSETS<= /span>){A20}, + 11922374 {B21} (b2) *(X2,=3D -15= .19280% {App1 C15}: X2=3DSG & A/SALES)= {A21}, + 587854.8 {B22} (b3,) * (X3, =3D 1079.15491% {D15}:X3=3DNET SALES/EMPLOYEES) {A22} + Residual ... 33291.63 {C43}.
SUMMARY, CONCLUSIONS AND IMPLICATIONS FOR F=
UTURE
STUDIES & RESEARCH
This study develops a “finger print&#=
8221;
definition for a specific class of frauds, with a specific rate, for a spec=
ific
industry. The regression parameters define parts of the fraud finger print.
Some of these parameters include such variables as the upper and lower boun=
ds
of the coefficients of the best fraud predictor variables, and its statisti=
cal
significance. This finger prints define Expert System (ES) rules for a Case
Based Reasoning (CBR) Knowledge Base (KB). The CBRKB should eventually cont=
ain
all the possible combinations of fraud finger prints for each Standard Indu=
stry
Classification (SIC) code. We also define the fraud rate of over or under
statements of financial accounts, and its time periods, as well as its amou=
nts
and their pattern, since they may affect the fraud finger print. Such finger
prints together with other fraud characteristics should help an ES diagnose
fraud patterns by benchmarking a suspicious company against its peers. The
resulting anomalies will fire the ES rules that will help trace the source =
of
the fraud.
This study finds the Best Predictor Percent=
age
Differenced Financial Ratios (BP%DFR) and the most effective in constructin=
g a
fraud cost model. These BP%DFR variables are estimating the Simulated Fraud
(SF) Dollar value, by building a regression model. This is case study of one
industry and one company but it may also apply to other companies and other
industries. In fact, our main ultimate objective is to deploy this model on=
ly
in cases where we have at least one other comparable company. So, even thou=
gh
we are using only a single focus company in the initial construction of the
model, future studies will focus on testing the model with more than one
company.
We decompose the fraud values into Fixed Fr=
aud
(FF), Variable Fraud (VF), and their sum Mixed Fraud (MF), building a regre=
ssion
model for each component separately, and for all of them combined. This
redundant computational effort is a control mechanism that we deploy to ens=
ure
accuracy, and trap errors. Thus, we develop a Cost Function (CF) for each f=
raud
component, FFCF, VFCF, & MFCF. These redundancy controls ensure reliabi=
lity
and accuracy.
We Simulate Fraud (SF) in= one account, which is the primary (1st) account. We balance this 1st account against the 2nd account. This balance equalizes the debit and the credit. FF behave different= ly than the VC. The FF remains f= ixed starting from the 1st period, to the last period. We have calculated the FF= as the higher of the 1st primary SF account VF rate or 1% of Net Sales, (which ever is higher). Add the FF to the VF and you get the MF for the last year.= Add this MF to the initial balance of the 1st account, real Dollar value balanc= e, for the last period, and you get the phony, fraudulent, balance of 1st acco= unt. Likewise, for the 2nd balancing account, start with the real balance, combi= ne the SF, and you get the final fraudulent value of the 2nd balanc= ing SF account.
We difference (deduct) the
Fraudulent, Post SF account balance, from the, real, Pre SF account balance,
redundantly calculating the SF Dollar values. Since we knew the SF Dollar v=
alue
apriori (before calculating it), recalculating the fraud is redundant, but
helps us confirm accuracy and reliability.=
Likewise, we difference all the Disclosure™ financial ratios
(%DFR). Regressing the SF Dollar Values on the %DFR enables us to discover =
the Best (highest R Square) Predictor Percentage
Differenced Financial Ratios (BP%DFR). In this case, these BP%DFR estimate =
the
SF Dollar values. We identify, explain and apply these BP%DFR variables.
Ultimately,
we intend to supplement existing Expert Systems (ES) software that can
discriminate between fraudulent and fraud free companies, but cannot pinpoi=
nt
the amounts of the fraud, and the its accounts. Integrating such a model into thes=
e ES,
will extend their abilities beyond current technology. Such extensions will
enable the ES to quantify the fraud and flag its sources (accounts). The
supplementary nature of model we develop explains its limitations. This mod=
el
is not for discovering fraudulent companies, since it will be redundant to =
the
existing ES. Thus, we optimize the current models for a relevant range that=
is
outside the vicinity of the origin, where both the %DFR and the Fraud appro=
ach
a value of greater than zero.
Therefore, we hypothesize that the intercept will be statistically
insignificant about the origin, which it turns out to be. In contrast to the
insignificance of the Intercept, we hypothesize that the entire model will =
be
significant, as well as at least one %DFR. Indeed, the ANOVA (Analysis Of
Variance) conforms our expectations, rejecting the null hypothesis that the
model is insignificant and that the regression coefficients are equal to
zero.
Since we intend to use it ultimately on a fraud suspect (company the ES has identified as a likely fr= aud reporting company) we refer to that company as the Focus company. In contra= st, the industry or competitors’ ratio averages constitute a fraud free P= eer Review Group (PRG). We calculate the %DFR as the ratios of the Focus less t= he Peers divided by the Peers. This assumption that fraud is the single differ= ence between the Focus and the Peers highlights the limitations of this approach. These limitations exclude heterogeneous industries where companies are not = very similar, as well as countries where fraud (bribes, etc.) is part of doing business, and GAAP/S (Generally Accepted Accounting Principles and Auditing Standards) are rare. It may be possible to apply these models to one or more companies within such countries, but not to companies of which one is from = such a country while the others are not from such countries. The confounding variables that emerge from the significant differences among the countries = may violate the assumptions of this model, invalidating the results.
We search for the ratios that will maximize=
the
correlation between the %DFR, the predictor variable(s) and the SF, the
predicted variable. We find that these ratios can help in auditing, detecti=
ng
and deterring fraud, as well as help develop rules for the Analytical Revie=
w ES
software (ARES). We apply these %DFR to one company, but design it to be mo=
re
generic, so we can expand it to other companies and industries. We develop a
case study of auditing this Focus company and its industry, where we simula=
te a
known rate of fraud. Our unique contribution to the existing literature is
going in ‘reverse’ to traditional auditing and accounting proce=
ss.
Instead of going from the individual transactions to the financial statemen=
ts,
we go in reverse, from the financial statements trying to reconstruct
fraudulent entries. Hence, is the concept Reversed Accounting Theory (RAT).
RAT, much like reversed engineering, goes in reverse to the common sequence=
of
work. We test the hypothesis that RAT can construct a statistically signifi=
cant
fraud function.
This study simulates fraud (SF) in the fina=
ncial
statements. This Simulated Fraud (SF) combines Variable Fraud (VF) and Fixed
Fraud (FF) into a Mixed Fraud (MF). We regress, correlate, & decompose =
such
a MF into its VF & FF components. We compare accounts, ratios, periods
& companies. We regress a Focus company with Simulated Frauds (SF) on i=
ts
simulated Peer Review Group (PRG) average (the Focus company itself, assume=
d to
be without the same MF). Of course, it is possible that our assumption that
this company without our SF does not have any such fraud is false. It is
possible that this company has already some of the fraud that we are
simulating, and our simulation just exaggerates the existing fraud. This
possibility stresses why it is important to repeat such studies on many
companies, and validate our results.
We apply RAT to Expert System (ES) software=
and
then test the hypothesis that RAT can decompose MF & help in flagging t=
he
MF riskiest financial ratios. Eventually our goal is to flag the riskiest
accounts that make up these ratios. Applying RAT, we decompose the MF into =
its
VF, & FF components. To decompose the MF, we regress it on sales. We de=
fine
the FF as regression’s intercept, and the VF as the regression’s
slope. Using the Least Square Regression we construct the MF function. We u=
se
actual sales (surrogates to forecasted sales as a function of serial date, =
or
earlier sales time series) as an independent predictor variable of the
dependent predicted variable, MF. This way, we construct a statistically
significant fraud model.
Applying
Activity Based Costing (ABC) theories to fraud, we develop the Activity Bas=
ed
Fraud (ABF) theory. ABF treats
fraud as a cost that we want to trace back to the activities that generate =
it,
much like any cost. Except th=
at for
fraud, unlike traditional costs, which we simply want to minimize, we want =
to completely
eliminate it, approximating a zero level fraud. To eliminate the fraud, we =
have
to estimate it (since it’s never reported voluntarily -- unlike other
costs), decompose it, leading to the activity sources. For that purpose we
deploy the Reversed Accounting Theory (RAT) to trace the fraud back to its
sources as a step to eliminate such fraud. Our unique contribution is going=
in
‘reverse’ to traditional auditing. Instead of going from the
individual transactions to the financial statements, we go in reverse, from=
the
financial statements trying to reconstruct fraudulent entries. Hence, we
develop the concept of RAT. Much like reversed engineering, RAT goes in rev=
erse
to the common sequence of accounting work. We test the hypothesis that RAT =
can
construct a statistically significant fraud function.
We deploy the Cost & Management Account=
ing
Theory (CMAT), and define VF, FF, & MF, as well as develop software
integrity controls. CMAT suggests that the slope of Mixed Cost (MC) regress=
ed
on the sales denotes Variable Cost Rate (VCR). Likewise, the intercept of MC
regressed on sales estimates the Fixed Cost (FC). Similarly, our $ SF regre=
ssed
on the sales has a slope and an intercept. This slope shows the Variable Fr=
aud
Rate (VFR), which is analogous to the VCR. At the same time, the intercept =
of
the SF on sales, shows the FF, analogous to the FC. RAT contends that whene=
ver
we are dealing with unknown fraud. Decomposing this fraud into its fixed and
variable components will be helpful in classifying its behavior and eventua=
lly
pinpointing its sources.
This study calculates risk for Analytical R=
eview
Diagnostics Controls for Automated Personal Computer (PC) Expert Systems (A=
RES)
based RAT. We integrate RAT, and the decomposition of this MF into the desi=
gn
of ARES, demonstrating some rules, screens, and reports that will result fr=
om
applying these ideas. By scanning financial statements and their ratios, and
regressing them against the peers averages the ARES could fire some rules.
These rules will quantify the risk of the likelihood of over or under stati=
ng
balances of accounts, helping auditors plan their audit. We could use such =
risk
measures to allocate audit time to accounts.
Likewise, such fraud coefficients help the =
ARES
prioritize its output and rate its opinions (related to problems in differe=
nt
accounts: phony sales or receivables) from the most to the least risky
accounts. Thus, the ARES could minimize over loading the user with reports =
that
they cannot process. This way the users can limit the ARES to report only on
the top most risky accounts and/or ratios, or pages of reports. This will
relieve the users from evaluating hundreds of accounts and ratios. This is
especially important when the users are not knowledgeable, do not have the
resources to conduct such an evaluation, or cannot set apriority the
materiality level. This could facilitate the deployment of ARES for
inexperienced managers and auditors.
We test the quality of this software using =
known
results, by using the company data as a surrogate to peers' standards and
numerous redundant calculations. We then use the unexpected differences as
internal controls to detect and correct such errors. Finding these errors
constitutes integrity controls and helps identify bugs. This is Total Quali=
ty
Management (TQM) methodology and approach, using industry based bench-mark
performance standards.
* We omitted additional appendices, tables,
glossaries, program code listings and screen shots to save space. We will provide such materials upon
written request.
&=
nbsp; &nbs=
p; &=
nbsp; &nbs=
p; &=
nbsp; &nbs=
p; &=
nbsp; =
REFERENCES=
=
=
REFERENCE WEB SITES AND QUOTES
Oliver Biggadike and Shannon D. Harrington, on Sept. 8 (Bloomberg) "Investors may be forced to settle contracts protecting mo= re than $1.4 trillion of Fannie Mae and Freddie Mac bonds against default" and Oct. 10 (Bloomberg)
http://www.bloomberg.com/apps/news?pid=3D20601087&a= mp;sid=3Daa6nmsv7BakE&refer=3Dhome "Lehman Credit-Swap Auct= ion Sets Payout of 91.38 Cents"
http://lawprofessors.typepad.com/securities/news_stories/index.html , August 23, 2007 “Rumors of a Sale of Bear Stearns"
http://msnmoney.bran= d.edgaronline.com/EFX_dll/EDGARpro.dll?FetchFilingHTML1?ID=3D5823375&Se= ssionID=3D5RgcWZDBP11rCl9 "Structured credit derivatives"
Matthew Monks, September 9, 2008, (mmonks@financialwee= k.com fw_editor@financialweek.com)
http://www.financialweek.co= m/apps/pbcs.dll/article?AID=3D/20080909/REG/809099979/1036 “Mortgage bailout could trigger massive credit default swap settlement Payouts"
PricewaterhouseCoopers LLP, Chartered Accountants and Registered Auditors, London "Report of Independent Registered Public Accounting Firm To the Board of Directors and Shareholders of Barclays Bank= PLC
Peter Viles, peter.viles@latimes.com, May 2007, http://latimesbl= ogs.latimes.com/laland/mortgages/ "Blame game: KPMG accuse= d of lax auditing of New Century."
=
=
REFERENCES
ON FRAUD, DAMAGES, DETECTION EXPERT SYSTEMS (EI) & DISCLOSURE
American Institute of Certified Public Accountants (AICPA), “The Auditor’= ;s responsibility to detect and report errors and irregularities”, ST= ATEMENT ON AUDITING STANDARDS NO. 53, auditing standard board, New York, 1988.<= o:p>
Berton,
Lee, “Auditors Face Stiffer Rules for Finding, Reporting Fraud at Cli=
ent
Companies”,
WALL STREET JOURNAL,
February 5, 1996, A2.
Coates,
P., and Fant, L., “A neural network approach to forecasting financial
distress”, THE JOURNAL OF BUSINESS FORECASTING, Winter 1991-2,
9-12.
Fanning,
Kurt; Cogger, O. Kenneth;
Srivastave, Rajendra “Detection of Management Fraud: A Neural
Network Approach” INTELLIGENT SYSTEMS IN ACCOUNTING, FINANCE AND
MANAGEMENT Vol. 4:113-126
(1995)
Fuerman,
R.,” The accounting profession’s litigation crisis”, O=
HIO
CPA JOURNAL, October 1992, 39-40.
Elgin,
P., “Huge liability judgments pressure CPAs to raise prices”, <=
u>CORPORATE
CASHFLOW, 13, July 1992, 12-14.
O’Mally,
S., “Legal liability is having a chilling effect on the auditor’=
;s
role”, ACCOUNTING HORIZONS, 72, June 1993, 82-7.
Palmrose,
Z., “Trials of legal disputes involving independent auditors: some
empirical evidence”, JOURNAL OF ACCOUNTING RESEARCH, 21,
Supplement, 1991, 149-185.
Pincus,
K., “The efficacy of red flag questionnaire for assessing the possibi=
lity
of fraud”, ACCOUNTING ORGANIZATIONS AND SOCIETY, 14, 1987, 153=
-63
St
Pierre, K. and Anderson, J., “An analysis of the factors associated w=
ith
lawsuits against public accountants”, THE ACCOUNTING REVIEW, 5=
9,
2, 1984, 242-63.
References on Activity Base Cost / Fraud (ABC/F), Short and Long Ru= ns, & Regression
Datar, Srikant; Gupta, Ma= hendra, “Aggregation, specification and measurement errors in
product costing”, Accounting Review , V= ol: 69 Iss: 4 , Oct 1994, 567-591.
Groth, John C; Kinney, Mi= chael R, “Cost management and value creation”, Management Decision, Vol: 32, Iss: 4, 1994 p: 52-5= 7.
Hartnett, Neil; Lowry, Jo= hn, “From ABC to ABM”, Australian Accountant, Vol: 64 Iss: 2 , Mar 1994 p: 28-32
Hartnett, Neil; Lowry, Jo= hn; Luther, Robert, “Is ABC feasible for external reporting?” Accountancy, Vol: 113 Iss: = 1209, May 1994, p: 74
Holmen, Jay S, “ABC=
vs.
TOC: It's a matter of time”, Management Accounting, Vol: 76
Macintosh, Norman B, “Management accounting's dark side: Part I”, CA Magazine, Vol: 127 Iss: 7 Sep 1994, p: 40-45
Mak, Y T; Roush, Melvin L, “Flexible budgeting and variance analysis in an activity-based costing environment”, Accoun= ting Horizons , Vol: 8 Iss: 2,= Jun 1994, p: 93-103
Pattison, Diane D; Arendt= , Carrie Gavan, “Activity-based costing: It doesn't work all the time”,<= /p>
Management Accounting<= /u>, Vol: 75 Iss: 10, Apr 1994, p: 55-61.
Sheu, Chwen; Wacker, John= G, “A planning and control framework for non-profit humanitarian organizations”, International Journal of Operations & Producti= on Management, Vol: 14 Iss: 4, 1994 p: 64-78
Smith, Malcolm, “Bo= ttleneck management”, Management Accounting-London, Vol: 73 Iss: 3 Mar 1995 p: 26-28.
Industry & Company Refere=
nces
REFERENCES OF BARCLAYS= AND BANKING INDUSTRY
Barclays(a), “President’s letter”, 10-K Annual Report, 12/31/94=
Barclays(b), “Manag= ement discussion”, 10-K Annual Report, 12/31/94
Five Paces Inc. Home Page=
Helfer, Ricki, The FDIC Quarterly Banking Profile, 12/31/95
C:\TEMP\FRAUD_BA.DOC Used: 11/23/95 12:57 PM Edit Time: 3 RevNum : 3 AR 1 Of 17