Neural networks, which are used to develop applications capable of "learning” from data over time, have been used with varying degrees of success in the stock market for several years. One developer, however, is now attempting to apply neural network technology to help develop accurate prepayment models and prices for mortgage-backed securities. It may, in fact, be one of the most amenable to neural networks' unique capabilities.

Baltimore-based Neuristics, a software research and consulting firm, is doing groundbreaking work in this area. The idea of applying neural networks to the valuation of mortgage-backed securities came out of Neuristics' original business, which centers around the application of neural networks and related technologies to credit scoring. The credit rating agencies—Experian (formerly TRW), TransUnion and Equifax—provide raw data on countless consumers. For these data to be useful, however, they must be organized into meaningful segments representing consumers who are likely behave in the same fashion. Then, it's necessary to find a way to use these data to help model how these segments will behave in the future. Neuristics has successfully applied neural networks and other artificial intelligence techniques to this problem, using data from the major credit bureaus in order to predict customer payment patterns.

Those same credit bureaus provide credit scores for mortgage holders. While the scores are sometimes used in valuing mortgage portfolios, they are typically not used as parameters in prepayment models. It occurred to developers at Neuristics that these credit data might be used as an additional predictor of prepayment rates. For example, most prepayment models do not adequately account for the segment of mortgage holders who will default or fall behind on their payments. Through the application of neural networks and the use of credit data, it may be possible to identify delinquent payers.

MINT and TIBCO, better together? The middleware market, although still quite young, is becoming increasingly specialized. Consider the recent alliance between MINT and TIBCO, which will allow users to combine the best features from both MINT's financial message transformation system and TIBCO's Enterprise Transaction Express (ETX) and TIB/Rendezvous products. In this configuration, TIBCO's products will provide the horsepower and manage the flow of information among different systems. MINT's transformation system will "translate” information into whichever format its eventual destination system requires. This could include third-party payment, clearing and settlement organization such as SWIFT, CREST, CHAPS and FIX. MINT's flexibility should allow users to introduce new message types quickly, and TIBCO's robust architecture should allow for fast, reliable and guaranteed message delivery. The alliance may indicate the start of a new direction in how users shop for middleware. Instead of selecting a single, generic middleware package, it may now be appropriate to piece together a best-of-breed middleware, which incorporates modules from two, three or more vendors. For more information about these companies, see: www.mintech.com and www.tibco.com.

Similarly, prepayment models are famous for their short life span. As soon as intrinsic market conditions change, prepayment models become obsolete. This phenomenon was perhaps most evident when mortgage brokers entered the market in large numbers earlier in the decade, actively encouraging mortgage holders to refinance and thus throwing existing prepayment assumptions out of whack. By applying neural networks to the generation of prepayment rates, it may be possible to identify market conditions in which it appears that the "old” rules of the road do not apply. This is because neural networks are data-driven, constantly updating their predictions—or predictive models—as new information comes in. "While neural networks and other advanced nonlinear techniques can better predict market behavior, their stability can rarely be trusted long enough for profitable use,” says Andy Krause, CTO for Neuristics. "This is because the underlying paradigms often change before they can be modeled, or remodeled, and applied. To combat this, Neuristics has developed proprietary data compression technologies that can alert users to paradigm shifts in real time to signal when a model is usable, when it is untrustworthy and when it requires retraining.”

In addition to their work in neural networks, Neuristics is also developing a data compression utility that will have applications for mortgage-backed securities traders and investors. When deciding whether or not to purchase a particular group of mortgages, many investors will consider credit data in order to help ascertain their collective worth. Typically, because raw credit data are so unwieldy, these investors will select data from one of the three major credit-scoring providers. With Neuristics' data compression utility, however, it will be possible to "clean” data provided by all three services, group this information into relative "domains” representing similar groups of mortgage holders, and then build dynamic relationships among these domains. In addition, the compression utility is able to reduce the quantity of data that need to be analyzed without compromising their meaning.

Although none of these products is on the market yet, it may represent the "next generation” for neural networks and other AI technologies.

Opening The Box

A cooperative development effort that actually produces usable software? Will wonders never cease?

Eight years ago, Financial Technologies International (FTI) assembled an international consortium of banking and trading organizations (including ABN AMRO, Cedel Bank, Fortis-MeesPierson, Toronto-Dominion Bank-Greenline Brokerage, The Northern Trust, and Wilmington Trust), that embarked upon an ambitious, $130 million development project. The goal was to build an enterprise-wide, multi-instrument, multicurrency accounting and back-office system.

Fast forward to July 1997. FTI has announced the availability of The Box Universal Financial Server, a real-time risk management, reporting and analytic system, which is one of the byproducts of this intensive development effort. Currently, The Box is in beta testing at various financial institutions; by the end of the year, it will be available for purchase.

Any large development effort involving multiple "cooks” naturally attracts skepticism. Many such efforts get so bogged down in internal politics that they start with a bang and then quietly disappear with nary a whimper. According to FTI's claims, however, it appears that The Box has promise—at the very least—as a firm-wide information-reporting tool.

By constantly querying local bank and market data systems, The Box maintains the consolidated current financial position of a customer, book, trader, counterparty, department, location or enterprise. It updates currency balances, securities positions, gain/loss and cash flows in real-time. It then stores this information, with the related transactions and market data, on a continuously accessible database that is structured to enable information requests from a wide range of unpredictable perspectives. This database serves as the single source of information for real-time web browser queries, as well as risk-management, compliance, audit, reporting, lending, borrowing, "one-call” customer service and other data-intensive applications.

According to a spokeswoman for FTI, much of the information provided by The Box typically has not been created in time to help active customers and financial managers. If produced at all, the information exists only on account statements or locked within a static warehouse, and is not available until after a batch reporting cycle. Even then, consolidating the desired information typically requires considerable manual effort by the user. By then, the time value of having the information has passed.

"Financial consolidation cycles range from 24 hours to 30 days, which today is too long to wait for actionable financial information,” says Charles J. "Chuck” Lewis, FTI Chairman and CEO. "By providing consolidated balances and projections in real time and on a comprehensive database, The Box bypasses the risk inherent in entering into financial commitments and making other decisions based on old or partial information. By also storing transaction and market details, The Box becomes a straight-through processing hub at the heart of an efficient information creation and transfer process—initiating multiple customer service, operational, administrative and financial management processes without human intervention or the delays caused by batch systems.”

If that wasn't enough, the buzz on The Box is that it will sell for much less than the "typical” enterprise-wide financial reporting system, which, when you throw in real-time processing, can easily reach a price tag of several million dollars. Although the system is in beta—that never-never land where anything is possible—it's definitely worth a look.

For more information, see www.ftintl.com.

Who employs how many in the multibillion dollar derivatives systems wars?

In the Derivatives Strategy Guide to Derivatives Technology, published in our July–August issue, we asked software vendors to list how many people they employed and in what capacity. Warning: Vendor's headcount numbers may be as inflated as their product claims. Caveat emptor.

Number of Employees at Derivatives Software Firms

Firm	Financial Engineers	Programmers	Support Technicians	Marketers	Administrative/ Other	Total
CMG	400	500	500	15	2,585	4,000
Bloomberg Financial Markets	100	800	1,100	600	400	3,000
Barra	n/a	n/a	n/a	n/a	n/a	500
MPCT Solutions Corp.	100	100	70	10	30	310
SunGard Capital Markets	41	110	98	28	18	295
Infinity Financial Technology Inc.	8	70	37	50	36	201
SunGard Futures Systems	4	47	125	11	9	196
Wall Street Systems Inc.	30	55	82	2	18	187
Algorithmics	40	80	0	20	20	160
OMR Systems Inc.	41	50	13	11	35	150
Summit Systems Inc.	45	45	45	10	5	150
FNX Limited	15	38	36	11	15	115
C*ATS Software	n/a	n/a	n/a	n/a	n/a	112
GE Information Services	5	28	30	34	7	104
Renaissance Software Inc.	30	35	15	10	10	100
Firm	Financial Engineers	Programmers	Support Technicians	Marketers	Administrative/ Other	Total
Sailfish Systems Ltd.	3	23	54	1	16	97
MUREX	20	25	42	0	7	94
Imagine Software	15	40	22	3	5	85
Mathsoft Inc.	1	40	5	4	30	80
Global Advanced Technology	5	35	13	5	20	78
DST Belvedere Financial Systems	9	16	14	5	33	77
ADS Associates Inc.	0	15	31	5	25	76
ATSM	4	40	15	10	5	74
Multinational Computer Models Inc.	10	20	25	5	10	70
Trema	4	15	15	5	26	65
Financial Software Systems	10	25	10	8	7	60
NeoVision Hypersystems Inc.	8	16	17	6	13	60
Inventure Ltd.	7	20	8	7	15	57
Brady Plc.	7	24	11	6	4	52
Open Link Financial Inc.	n/a	n/a	n/a	n/a	n/a	50
INSSINC	8	14	10	5	10	47
Monis Software Inc.	9	16	10	6	6	47
Firm	Financial Engineers	Programmers	Support Technicians	Marketers	Administrative/ Other	Total
FSD International	12	20	8	1	3	44
Boston Treasury Systems	3	14	15	3	5	40
NumeriX LLC	10	15	4	3	5	37
Platinum Treasury Systems Inc.	3	12	12	4	4	35
ZAI*NET	3	10	12	5	5	35
Tech Hackers Inc.	4	10	4	8	6	32
Investment Intelligence Systems Corp.	5	13	7	2	3	30
LOGIN SA	9	13	0	2	6	30
Timeris	n/a	n/a	n/a	n/a	n/a	30
Axiom Software Laboratories	3	8	7	3	4	25
Software Options Inc.	n/a	n/a	n/a	n/a	n/a	25
Advanced Risk Management	10	6	3	3	2	24
Solutions Pte.Ltd.(ARMS) Benton Associates Inc.	2	7	9	3	1	22
Triple Point Technology Inc.	0	9	7	1	5	22
LongView International Inc.	3	8	3	4	3	21
Firm	Financial Engineers	Programmers	Support Technicians	Marketers	Administrative/ Other	Total
CastleNet LLC	5	9	1	1	4	20
ITS Trading Systems Ltd.	2	5	4	2	7	20
Theoretics Mountain View Inc.	5	5	3	4	2	19
Derivative Solutions Inc.	4	3	6	3	1	17
Redpoint Software Inc.	2	5	2	5	3	17
Primo Systems Inc.	2	7	2	3	2	16
Financial Engineering Associates Inc.	6	2	2	2	2	14
Savvysoff	4	4	4	1	1	14
Savid International Inc.	1	4	4	3	1	13
Decision Software Inc.	6	0	4	1	1	12
DiRollo Jackman	2	5	2	2	1	12
Mamdouh Barakat Risk Management	3	3	2	2	2	12
FinaTech Ltd.	3	2	2	2	2	11
Intermark Solutions	2	2	3	3	1	11
TruRisk Inc.	2	5	2	0	1	10
Fxpress Corp.	1	2	2	2	1	8
Firm	Financial Engineers	Programmers	Support Technicians	Marketers	Administrative/ Other	Total
Super Computer Consulting	1	2	1	3	1	8
Trade Edge Solutions	1	3	2	1	1	8
Capital Market Technology Inc.	2	3	0	1	0	6
IRIS Integrated Risk Management AG	n/a	n/a	n/a	n/a	n/a	6
Lester Associates	0	5	0	0	1	6
A-J Financial Systems	1	2	0	0	1	4
Ian Merker and Co.Ltd.	2	2	0	0	0	4
Risk Management Software Inc.	1	1	0	0	0	2