USOO6349291B1 (12) United States Patent (10) Patent No.: US 6,349,291 B1 Varma (45) Date of Patent: Feb. 19, 2002 (54) METHOD AND SYSTEM FOR ANALYSIS, OTHER PUBLICATIONS DISPLAY AND DISSEMINATION OF FINANCIAL INFORMATION USING Hansson, B., and Persson, M. “Time Diversification and RESAMPLED STATISTICAL METHODS Estimation Risk”, Financial Analysts Journal, vol. 56, No. 5 (Sep./Oct. 2000): pp.55-62.* (75) Inventor: Samir Varma, Greenwich, CT (US) Gardner, G., and Stone, D. “Estimating Currency Hedge Rations for International Portfolios”, Financial Analysts (73) Assignee: Attractor Holdings LLC, Greenwich, Journal, vol. 51, No. 6 (Nov./Dec. 1995): start p. 6.* CT (US) Choong, N.K., and R. McLeod, Jr. “Expert, linear models, and nonlinear models of expert decision making in bank Notice: Subject to any disclaimer, the term of this ruptcy prediction: a lens model analysis”, Journal of Man patent is extended or adjusted under 35 agement Information Systems, Vol. 16, No. 1, (Summer U.S.C. 154(b) by 0 days. 1999): pp. 189–206.* Mitchell, M., and Stafford, E. “Managerial decisions and (21) Appl. No.: 09/489,364 long-term Stock price performance', The Journal of Busi ness, vol. 73, No. 3 (Jul. 2000): pp. 287-329.* (22) Filed: Jan. 21, 2000 Laster, D. "Measuring gains form international equity diver sification: The boofstrap approach”, journal of investing: (51) Int. Cl................................................. G06F 17/60 New York; vol. 7, No. 3, (Fall 1998): pp. 52–60.* (List continued on next page.) (52) U.S. Cl. ............................................ 705/35; 705/37 Primary Examiner Vincent Miller ASSistant Examiner-Daniel S Felten (58) Field of Search .............................. 705/35, 36, 37, (74) Attorney, Agent, or Firm-Kenyon & Kenyon 705/38 (57) ABSTRACT (56) References Cited The present invention provides a method and System for the U.S. PATENT DOCUMENTS Statistical analysis, display and dissemination of financial 4,541,115 A 9/1985 Werth .......................... 282/14 data over an information network Such as the Internet and 5,283,856. A 2/1994 Gross et al. .................. 395/51 WWW. The present invention utilizes resampled statistical 5,359,699 A * 10/1994 Tong et al. ................... 395/22 methods for the analysis of financial data. ReSampled Sta 5,598.439 A * 1/1997 Wagner ..... ... 375/326 tistical analysis provides a meaningful and reasonable Sta 5,774,878 A * 6/1998 Marshall ...................... 705/35 tistical description of financial information, which typically 5,790,177 A * 8/1998 Kassatly .... 348/13 escapes modeling using parametric methods (i.e. assump 5,893,069 A * 4/1999 White, Jr. ...................... 705/1 tions of a Gaussian distribution). The present invention 5,918,217 A * 6/1999 Maggioncalda et al. ...... 705/36 provides at least a GUI that provides functionality for user 5,946,666 A * 8/1999 Nevo et al. ................... 705/36 input of Statistical queries, a Statistical computation engine 6,003,018 A * 12/1999 Michaud et al. .............. 705/36 6,012,043 A * 1/2000 Albright et al. .............. 705/36 that performs Statistical analysis of financial data and a 6,064,662 A 5/2000 Makivic ...................... 705/36 graphical rendering engine that generates graphical display 6,064.984. A * 5/2000 Ferguson et al. ............. 705/36 of Statistical distributions generated by the Statistical com FOREIGN PATENT DOCUMENTS putation engine. According to one embodiment, the present invention employs a parallel processing architecture to JP 8-22.1573 * 8/1996 Speed generation of the resampled Statistics. JP 8-221574 * 8/1996 JP 10-512985 * 12/1998 31 Claims, 17 Drawing Sheets US 6,349,291 B1 Page 2 OTHER PUBLICATIONS Putnam, B. “Is the trend still my friend?” Global Investor, Kemp A. W. “Bootstrap Methods: A Practitioner's Guide', No. 66, (Oct. 1993): Start p. 21.* Biometrics, vol. 56, No. 1 (Mar. 2000): Start p. 317.* Caudill, S., and Holcombe, R. “Specification search and Fiellin, D., and Feinstein, A. “Bootstraps and jackknives: levels of Significance in econometric models”, Eastern New computer-intensive Statistical tools that require no Econonic Journal, vol. 25, No. 3 (Summer 1999): pp. mathematical theories” Journal of Investigative Medicine, 289-300. Deis, D., and Hill, R., “An application of the bootstrap vol. 46, No. 2, (Feb. 1998): start p. 22.* method to the Simultaneous equations model of the demand Berkowitz, J., and Kilian, L., “Recent Developments in and Supply of audit Services, Contemporary Accounting Research, vol. 15, No. 1 (Spring 1998): pp. 88–99.* Bootstrapping Time Series”, manuscript, Sep. 25, 1996.* Magnar, L., and Steinar, E., “Exact confidence intervals generated by conditional parametric bootstrapping”, Journal Michaud, R. The Markowitz Optimization Enigma: Is Opti of Applied Statistics (May 1999): pp. 447–459.* mized Optimal?, Financial Analysts Journal, (Jan./Feb. Coleman, M., Lynford, L., and James, F., “A newly inte 1989): pp. 31-42.* grated and dynamic approach to Securitized real estate risk analysis and valuation” Real estate Finance, vol. 12, No. 1 (Spring 1995): Start p. 68.* * cited by examiner U.S. Patent Feb. 19, 2002 Sheet 3 of 17 US 6,349,291 B1 . O L U.S. Patent Feb. 19, 2002 Sheet 4 of 17 US 6,349,291 B1 Preference Parameter 1-340(1) Preference Parameter 2-340(2) Preference Parameter N-1-340(N-1) Preference Parameter N-340(N) F/G 3 405 Investment ID - 410 InVestment Name - 415 InVestment Data' - 420 A/G 4 U.S. Patent Feb. 19, 2002 Sheet S of 17 US 6,349,291 B1 505 Rule D-510 Rule Object" - 515 F/G, 5a Function - 525 A/G 6f) U.S. Patent Feb. 19, 2002 Sheet 6 of 17 US 6,349,291 B1 Return Object N-1-625(N-1) Return Object N-625(N) F/G, 6a 625 Date - 630 Value - 635 F/G, 6f) U.S. Patent Feb. 19, 2002 Sheet 7 of 17 US 6,349,291 B1 705 Function prototype ID-710 Function Prototype Object - 715 A/G 7 Return Object N-1-625(N-1) Return Object N-625(N) A/G 3 U.S. Patent Feb. 19, 2002 Sheet 9 of 17 US 6,349,291 B1 - - - - - - -- - - - - - - - - -- - -- - -- - - - - - - - - m -- - - -- - - - - - - - - - - -- - - -- - - - - - - - Generate EDF, Fa(6) Bootstrap 935 - - tReplications e \.or er - ---e U.S. Patent Feb. 19, 2002 Sheet 10 of 17 US 6,349,291 B1 uest for Resampled 1005 Regtatistical Analysis Receive input Parameters -100 Request Aysis RSA FOm ReSampling Computation Finished 1025 U.S. Patent Feb. 19, 2002 Sheet 11 of 17 US 6,349,291 B1 Initiate Resampled 1105 1135 Statistical Method O BiaS Reguese yes Set SampleSpace 1115 PerformAlgorithm Bias initialization 1140 Calculate Memory 1117 Requirements. Request Memory Range in Shared 1145 Memory Area and Output Data To Be emory Area Jansgrmed O yes Transfer Selected Sample 1119 SpaceData from Raw Transform Data 1150 Financial database to Portion of Shared Memory 1160 Area M = Int(Q/N) Calculate Basic Statistics 6 h-1120 Initiate Parallel Processing 1165 For Range of Lag intervals 1125 Calculate AutoCorrelations 1170 (a) and Partial Parallel AN/ AutoCOrrelations PrOCeSSin Complete Determinelag Factor (N) 1130 That Minimizes AutoCorrelation (a) = Notify Back End Processor and TranSmit Memor 1175 Number of Consecutive Periods of Data to Pick at Address in Output Data One Time (N) Area Terminate. 1180 Initiation ROutine F/G 11 U.S. Patent Feb. 19, 2002 Sheet 12 of 17 US 6,349,291 B1 Initiate Parallel 1205 Processing Perform Load Balancing 1210 Analysis Request Parallel Process on r-1220 PrOCeSSOr 1-P 1230 All Parallel PrOCeSSes O Completed yes ASSemble Data Area Output Datar\-1240 Notify Control Server of 1250 Completion Initiate Parallel 1260 Processing F/G, 12 U.S. Patent Feb. 19, 2002 Sheet 13 of 17 US 6,349,291 B1 Initiate PrOCeSS 1305 Determine Output Address 1310 Range Set Random Start Point in 1320 Array 1360 1325 -/ 1365 Notify Server of COmpleted Replication-9 Build Bootstrap Sample "replication Pick N Consecutive Periods. Using Bias and Stole internporary Array COunt = COunt +1 Calculate Replication Calculate s(x"replication) S("replication) 1355 St0re s(x'replication) A/G 13 U.S. Patent Feb. 19, 2002 Sheet 14 of 17 US 6,349,291 B1 Initiate BiaS 1405 Separate SampleSpace into 1410 Up Day Set get Down Day Pick Random Number?, 1420 Between 0-1 1440 Y Pick Up Day Pick Down Day 1450 A/G, 14 P RÈ US 6,349,291 B1 S?S#E.38 £(6H[0Y12M]INS Q_2 F –– – –+ H- –] 1 –%9? ---|-------*----|---------%|,0Z |---------,----XOOO?OZ/------|-|-|-|-------%0. _-._ _•%68_ _%9 E O U.S. Patent Feb. 19, 2002 Sheet 17 of 17 US 6,349,291 B1 ·ØN SN N ZZZZZZZZZZZZZZZZZZZZZZZZZ ZZZ ZZZZZZZZZZZZZZ 22 NNNN &N NXSSS N ZZZZZZ VALKOZRNFTØGNET? NYNY YYY NOE Ø N. Y. Ya Ya Ya Ya Ya YYYY ZZZZZZZZZZZZZZZZZZZZZZZZ Y N ZZZZZZZ S00uen000 JOU000d US 6,349,291 B1 1 2 METHOD AND SYSTEM FOR ANALYSIS, Statistical description of financial information, which typi DISPLAY AND DISSEMINATION OF cally escapes modeling using parametric methods (i.e., FINANCIAL INFORMATION USING assumptions of a Gaussian distribution). RESAMPLED STATISTICAL METHODS The present invention includes a financial information network node that is coupled to an information network Such FIELD OF THE INVENTION as the Internet. The financial information network node The present invention relates to the area of electronic includes a front end Subsystem, a resampled Statistical information Systems. In particular, the present invention analysis engine ("RSAE) and a graphics rendering engine relates to a method and System for the delivery of financial (“GRE”). The front end Subsystem provides a graphical user information using resampled Statistical methods over an interface (“GUI”) that allows clients also coupled to the information network. information network to Submit requests for resampled Sta tistical analysis of various financial investments and receive BACKGROUND INFORMATION graphical display of the results. The RSAE performs resa mpled Statistical analysis of financial data in response to user Investors and financial analysts rely upon electronic infor 15 queries and incorporates routines to preserve temporal cor mation Systems for the delivery of accurate financial and relation in financial data, which necessarily provides more investment data and analysis in order to devise meaningful accurate analysis. In addition, the RSAE provides for user investment strategies. The growth of the Internet and World control of a number of parameters to Simulate various Wide Web (“WWW) highlights the potential for global financial environmental conditions. For example, according distribution of “real time' or “near real time' financial to one embodiment, the RSAE allows a user to simulate information and analysis. For example, a number of WWW either bull or bear market conditions by Setting a bias SiteS provide financial information to clients Such as inves parameter that controls a degree of randomneSS in the tors and financial analysts. resampling process. The GRE generates a graphical display However, conventional financial information Sites do not of statistical distributions generated by the RSAE. provide meaningful analysis tools to accurately analyze, 25 According to one embodiment, the present invention forecast and predict the behavior of financial markets. Con employs a parallel processing architecture to Speed genera ventional technology for delivery of financial information tion of the resampled Statistics. The parallel architecture is over information networkS Such as the Internet typically afforded by the nature of the resampling algorithm itself, allows users to track returns for various investments and which permits the financial data to be vectorized. This perform rudimentary Statistical analysis (e.g., computation parallel processing architecture provides at least two Sig of the mean and Standard deviation) for these investments. nificant advantages. First, the architecture permits the deliv However, these rudimentary Statistical functions are not ery and processing of financial data in compressed time useful to investors in forecasting the behavior of financial frames, which facilitates “real time” or “near real time” markets because they rely upon assumptions that the under Statistical analysis. In addition, the parallel computation lying probability distribution function (“PDF) for the finan 35 Scheme provides the ability to perform Statistical analysis on cial data follows a normal or Gaussian distribution, which is a large number of financial entities (e.g., a mutual fund or generally false. hedge fund) through a weighting process. The true distribution of returns for any financial market According to one embodiment of the present invention for (and thus of a trading strategy) is unknown. It is therefore 40 implementation on the Internet, a financial information site incorrect to rely upon a Statistical model based on assump is coupled to the Internet via a front end Subsystem including tions of normality (e.g., standard deviation). Typically, the a WWW server. The financial information site includes a PDF for financial market data is heavy tailed (i.e., the front end Subsystem, a RSAE and a GRE. In addition, the histograms of financial market data typically involve many financial information Site maintains a database of financial outliers containing important information). Thus, Statistical 45 data for any number of financial entities Such as companies, measures Such as the Standard deviation provide no mean mutual funds etc. The financial information site also main ingful insight into the distribution of financial data. tains a database of clients that have registered with the Providing reasonable methods for the analysis of financial financial information Site and desire to obtain Statistical market data is essential for investors. Reasonable Statistical analysis of financial data. analysis of financial data should at a minimum provide an 50 In order to perform a resampled Statistical analysis, a accurate assessment of potential financial risk and reward. query is received from a client via the front end Subsystem. However, conventional methods, which rely upon assump A client may specify a number of parameters including an tions of a Gaussian distribution, are dangerous to investors investment or investments (e.g., a portfolio) to be analyzed, because these analyses understate the true risk and overState a financial function, a Sample size, a period, a type of plot potential rewards for an investment or trading Strategy. 55 and a bias parameter, which controls the randomness of the Thus, this information is not generally useful and if relied resampling process. Based upon the parameterS Specified by upon promotes imprudent investment decisions. In general, the client, the RSAE performs a resampled Statistical analy the heavy tailed nature of financial data presents significant sis of relevant financial data. The GRE then produces a challenges in providing meaningful Statistical analysis. distribution plot based upon the output generated by the 60 RSAE, which is presented to the client via the front end SUMMARY OF THE INVENTION Subsystem. The present invention provides a method and System for According to one embodiment of the present invention, the Statistical analysis, display and dissemination of finan the RSAE performs at least three types of financial functions cial data over an information network Such as the Internet on financial data. A groSS rate of return function provides and WWW. The present invention utilizes resampled statis 65 analysis of the groSS rate of returns for an investment over tical methods for the analysis of financial data. ReSampled a specified time period. A maximum drawdown function Statistical analysis provides a meaningful and reasonable provides analysis of a maximum drawdown for an invest US 6,349,291 B1 3 4 ment over a Specified period. A monitor function provides grOSS rate of returns according to one embodiment of the analysis of a number of “up” and “down” days for a present invention. particular investment over a period of time. FIG. 16 is an exemplary plot of a resampled Statistical The financial information site also provides functionality analysis comparing two investment Strategies with respect to for Storing a set of client specified alert rules that are used maximum drawdown returns according to one embodiment to automatically monitor the behavior of investments based of the present invention. upon a resampled Statistical analysis process and notify FIG. 17 is an exemplary plot of a resampled statistical clients of the financial information site when the behavior of analysis comparing multiple investment Strategies with a particular investment violates a specified rule. respect to a monitor function according to one embodiment BRIEF DESCRIPTION OF THE DRAWINGS of the present invention. DETAILED DESCRIPTION FIG. 1 is a block diagram of a network architecture that illustrates the relationship between a financial information Although the embodiments described herein utilize the Site and a client according to one embodiment of the present 15 Internet and WWW, the present invention is compatible with invention. any type of information network public or private and thus, FIG. 2 is a detailed block diagram of a financial infor the embodiments described herein are not intended to limit mation site according to one embodiment of the present the Scope of the claims appended hereto. For example, the invention. present invention could be implemented using a private FIG.3 depicts the structure of a client record that is stored Intranet, local area network (LAN), metropolitan area net in a client database at a financial information site according work (MAN), wide area network (WAN) or even a wireless network to one embodiment of the present invention. FIG. 1 is a block diagram of a network topology that FIG. 4 depicts the structure of an investment record that illustrates is Stored in an investment database at a financial information the relationship between the Internet, a financial Site according to one embodiment of the present invention. 25 information embodiment site and various clients according to one of the present invention. Based upon queries FIG. 5a depicts the structure of an alert rule record that is submitted by clients, financial information site 119 performs Stored in an alert rules database at a financial information Site according to one embodiment of the present invention. resampled Statistical analysis of financial data and provides a graphical display of distribution results. Details of the FIG. 5b depicts the structure of a rule object record functionality provided by financial information site 119 are according to one embodiment of the present invention. described below. FIG. 6a depicts a data Structure for Storing financial data Clients 105a-105c communicate with financial informa in a financial database according to one embodiment of the tion site 119 via Internet 114. According to the embodiment present invention. depicted in FIG. 1, financial information site 119 is coupled FIG. 6b depicts a data Structure for Storing a financial 35 to Internet 114 via T1 line 130b. Client 105a illustrates a return according to one embodiment of the present inven typical narrowband client coupled to Internet 114 via a tion. dial-up connection described in more detail below. Client FIG. 7 depicts a data structure for storing a function 105b illustrates a typical broadband client coupled to Inter prototype in a function database at a financial information net 114 via a cable modem. Client 105c illustrates a corpo Site according to one embodiment of the present invention. 40 rate client that is coupled to Internet via T1 line 130c and FIG. 8 depicts a data structure for storing plot information server 151. Corporate client 105c includes three network in a plot database at a financial information site according to nodes 171a–171c that share bandwidth on Ethernet 161. one embodiment of the present invention. Although FIG. 1 illustrates three clients (105a-105c), it is FIG. 9a (reprinted from Efron and Tibshirani) depicts the to be understood that financial information site 119 may underlying theory of the bootstrap method. 45 serve any arbitrary number of clients 105 limited only by the FIG. 9b depicts a procedure for performing a bootstrap processing power and bandwidth available. method to generate a distribution of bootstrap replications As illustrated in FIG. 1, client 105a communicates with according to one embodiment of the present invention. financial information site 119 via personal computer 112a, FIG. 10 is a flowchart of steps for performing a resampled modem 115a, POTS telephone line 117 and Internet service analysis of an investment and generating a graphical output 50 provider 120a. Internet service provider 120a includes according to one embodiment of the present invention. modem bank 121 and router 135a that routes packets received from modem bank 121 onto Internet 114 via T1 line FIG. 11 is a flowchart that depicts a set of steps to initiate 130a. Packets are routed over Internet 114 to client gateway a resampled Statistical analysis of financial data using a server 14.0a at financial information site 119 via T1 line parallel processing architecture according to one embodi 55 13Ob. ment of the present invention. Client 105a utilizes personal computer 112a to navigate FIG. 12 is a flowchart of a parallel processing control Internet/World-Wide-Web (WWW 114 via browser software process according to one embodiment of the present inven (not shown) and display device (not shown). The browser tion. Software permits navigation between various file Servers FIG. 13 is a flowchart of a set of steps for performing a 60 connected to Internet 114, including client gateway Server reSampled Statistical analysis according to one embodiment 14.0a at financial information site 119. The browser Software of the present invention. also provides functionality for rendering of files distributed FIG. 14 is a flowchart of a set of steps for performing a on the Internet (i.e., through plug-ins or Active X controls). biasing procedure according to one embodiment of the In order to transmit data to financial information site 119, present invention. 65 personal computer 112a transmits signals through a dial-up FIG. 15 is an exemplary plot of a resampled statistical connection utilizing modem 115a. Modem 115a performs analysis comparing two investment Strategies with respect to modulation of digital Signals generated by personal com US 6,349,291 B1 S 6 puter 112a onto an analog carrier Signal for transmission requests to RSAE for resampled Statistical analysis of finan over the public Switched telephone network (“PSTN) (not cial data and Submits requests to GRE for graphical render shown). Modem 115a also performs demodulation of signals ing of output generated by RSAE. Back end server 140b received over local lines (e.g., 117) from the PSTN extract communicates with control server 140c at RSAE 139, ing digital byte codes from a modulated analog carrier. 5 graphics rendering server 140e at GRE 149 and SMTP Signals are received at ISP 120a, which is connected to server 140fat front end 1 subsystem 29. the PSTN through modem bank 121. Digital IP packets are RSAE 139 includes control server 140c, parallel process then transmitted via Internet 114 and various routers (not control server 140d, parallel processors 112a-112e (each shown) to WWW server 140a. IP packets are also transmit including local respective cache 112a1-112e 1), financial ted in the reverse direction from WWW server 14.0a to 10 database 150d, investment database 150e, function database personal computer 112a. 150f shared memory area 160a and output data area 160b. Client 105b is coupled to Internet 114 via a broadband Note that RSAE 139 depicted in FIGS. 1-2 utilizes a parallel cable connection. In particular, personal computer 112b processing architecture. This parallel Scheme is merely transmits packets via cable modem 115b to ISP 120b where 15 exemplary and is not intended to limit the Scope of the the packets are routed over Internet 114 to client gateway claims appended hereto. Other embodiments may not rely server 140a. Packets from financial information site 119 upon a parallel processing architecture at RSAE 139. traverse a reverse path to client 105b. Similar to client 105a, Control server 140c provides communication functions client 105b utilizes browser Software to navigate Internet between back end server 140lb and RSAE 139 and controls 114 and WWW. the overall operation of a resampled Statistical analysis Corporate client 105c includes network nodes 171a–171c, process. Control Server 140C is coupled to parallel proceSS which are coupled to Internet via Ethernet 161, server 151 control server 140d and shared memory area 160a. Shared and T1 line 130c. Network nodes 171a–171c may commu memory area 160a Stores Sample data for financial invest nicate with financial information site 119 via Ethernet, ments currently being analyzed by RSAE 139. As described server 151, T1 line 130c, Internet 114 and T1 line 130b. 25 in detail below, control server 140c receives requests for Similar to clients 105a-105b, it is assumed that users at parallel processing computations from back end Server network nodes 171a–171c utilize browser Software to navi 140b, performs certain initialization functions, loads appro gate Internet 114 and WWW. priate data into shared memory 160a and forwards these The specific nature of clients 105a-105c and the methods requests to parallel process control Server 140d for perfor through which they are coupled to Internet 114 depicted in mance. Control server 140c then waits for a completion FIG. 1 are merely exemplary. The present invention is Signal from parallel process control Server 140d and pro compatible with any type of Internet client and/or connec vides the output results to back end server 140b for further tion (broadband or narrowband). In general, it is to be processing (e.g., graphical rendering via GRE 149). understood that clients 105 may connect to Internet 114 Control server 140c is also coupled to financial database using any potential medium whether it be a dedicated 35 150d, investment database 150e, function database 150f and connection such as a cable modem, T1 line, DSL (“Digital shared memory area 160a. Financial database 150d Subscriber Line”), a dial-up POTS connection or even a (described in more detail below) stores financial Sample data wireleSS connection. relating to particular investments. Investment database 150e FIG. 2 is a detailed block diagram of a financial infor (described in more detail below) Stores financial data regard mation site according to one embodiment of the present a ing investments for which clients may be interested in invention. Financial information site 119 includes front end performing resampled Statistical analysis (i.e., Stocks, subsystem 129, RSAE 139, GRE 149, back end server 140, mutual finds, etc.). Function database 150f (described in client database 150g and alert rules database 150c. more detail below) stores function prototypes for functions Front end subsystem 129 includes client/gateway server to be performed on financial data. 140a, which is coupled to GUI database 150a. Front end 45 Parallel process control server 140d is coupled to parallel Subsystem 129 provides a GUI, which allows clients 105 to processors 112a-112e and output data memory area 160b. transmit information to and receive information from finan Parallel processors 112a-112e, which are each coupled to a cial information site 119. According to one embodiment respective local cache 112a1-112e 1 and shared memory GUI database 150a stores HTML (“Hypertext Markup area 160a, perform resampled Statistical analysis of Sample Language') code (i.e., WWW pages) relating to various so data stored in shared memory area 160a (i.e., resampled Statistical computations). Parallel process control server information and functions provided by financial information site 119. For example, GUI database 150a may store a 140d (described in more detail below) orchestrates and HTML “home page” for financial information site 119 or controls parallel computation processes running on parallel HTML pages including forms, which allow the input of data processors 112a-112e. In particular, parallel process control at financial information site 119. 55 Server 140d requests initialization of resampled Statistical Front end subsystem 129 also includes SMTP (“Simple analysis of data stored in shared memory area 160a from Mail Transport Protocol”) server 140f SMTP server 140f individual processors 112a-112e. Upon completion of all performs transmission of e-mail messages to clients 105 parallel processes running on processors 112a-112e, parallel associated with financial information site 119 in order to process control server 140d retrieves the results stored in provide notification regarding various events (as described 60 local caches 112a1-112e 1 and Stores the aggregate data in in more detail below). output data area 160b where it can be processed further (e.g., Client gateway server 14.0a communicates with back end in GRE 149). server 140b, which controls and orchestrates the large-scale GRE 149 performs graphical rendering (e.g., plots) of processing of data at financial information site 119. In output data generated by RSAE 139. GRE 149 includes particular, back end server 140b handles responses to 65 graphics rendering engine Server 140e, which is coupled to requests from clients 105 for resampled statistical analysis plot database 150b. As described in detail below, plot of investments. For example, back end server 140b Submits database 150b Stores data regarding the rendering and for US 6,349,291 B1 7 8 matting of distribution plots generated by graphics rendering information site 119 executes a process to notify clients if a engine server 140e. threshold percentile of a resampled Statistical distribution is Back end server 140b is also coupled to client database either below or above a current value of a financial event, 150g and alert rules database 150c. Client database 150g indicating that the event is unlikely. For example, a client Stores information related to clients that have registered with 105 may desire to be alerted if the gross rate of returns for financial information site 119. Alert rules database 150c a specified investment over a 200-day period assumes an Stores data pertaining to client specified rules for alerting improbable value. In this case, each day (or at a frequency clients to near real time behavior of investments. According calculates the aactual specified by client 105), financial information site 119 groSS rate of returns for the investment to one embodiment of the present invention, clients are over the last 200 days. Then, financial information site 119 alerted to rule violations by e-mail via SMTP server 140f executes a resampled Statistical process to evaluate the groSS which is also coupled to back end server 140b. rate of returns for 200-day periods described in detail below) FIG.3 depicts the structure of a client record that is stored to determine whether a percentile value of the distribution is in a client database 150g at a financial information site 119 above or below the current value. If so, the current value is according to one embodiment of the present invention. Each highly unlikely and the client 105 is notified via e-mail. client record 305 includes client ID field 310, client pass 15 Each alert rule record 505 includes rule ID field 510 and word field 315, portfolio pointer field 320, alert rules* rule function object pointer field 515. Rule ID field 510 pointer field 325, e-mail address field 330, billing parameter Stores a unique 32-bit integer value pertaining to an alert field 335 and preference parameter fields 340(1)–340(N). rule, which is used for identification purposes. Rule function Client ID field 310 stores a unique 16-byte character array object pointer field 515 stores a reference to a rule object or pointer to a character array of a client that has registered (described with reference to FIG. 5b) relating to the occur with financial information site 119. Client password field rence of a financial event for which a client desires notifi 315 Stores a unique 16-byte character array or pointer to a cation. character array of a password associated with a client 105. FIG. 5b depicts the structure of an alert rule object Clients 105 may establish a client ID and password upon 25 according to one embodiment of the present invention. Each registration with financial information site 119. rule alert rule record 507 includes investment ID field 520, Portfolio pointer field 320 stores a pointer to a linked list function field 525, periods field 530, operator field 535, of investments that a client 105 has selected for tracking percentile value field 540, Sample size field 545 and repli using financial information Site 119. According to one cations field 550. Investment ID field stores a 32-bit integer embodiment of the present invention, each link in the linked value identifying an investment, as described below with list Stores an identifier of an investment entity as described respect to FIG. 6a. Function field 525 stores a 32-bit in more detail below. Alert rules* pointer field 340 stores a function ID of a function record as described below with linked list of alert rule record IDs (discussed in more detail respect to FIG. 7. Periods field 530 stores a number of below) that specify particular financial alert rules that are periods (i.e., days) for which the client 105 desires to monitored by financial information site 119 and associated 35 evaluate the investment. Operator field 535 stores a 4-bit with individual clients 105. These rules are used to notify field indicating an operator Such as < or'>. Percentile individual clients 105 of the occurrence of particular events value field 540 Stores an integer representing a percentile they wish to follow based upon a resampled Statistical value. Sample size field 545 stores a 32-bit integer value analysis of financial data. E-mail address field 330 stores a representing a Sample size for which to conduct a resampled 32-byte character array or pointer to a character array of an 40 statistical analysis. Replications field 550 stores a 32-bit e-mail address of a client 105. Billing parameter field 335 integer value representing a number of replications to per Stores a pointer to billing object record that includes billing form in conducting a resampled Statistical process. information for a client 105. Preference parameters 340(1) AS described in detail below, the resampled process is -340(N) store preference parameters related to customiza conducted based upon parameters stored in fields 520, 525, tion functions associated with financial information site 119. 45 530, 545 and 550. Based upon operator filed 535, it is then FIG. 4 depicts the structure of an investment record that determined whether the distribution results for a resampled is stored in an investment database 150e at a financial Statistical proceSS above or below the current value exceed information site 119 according to one embodiment of the the percentile value stored in percentile value field 540. If so, present invention. Investments may represent Stocks, mutual the client is notified funds, etc. Each investment record 405 includes investment 50 FIG. 6a depicts a data Structure for Storing financial data ID field 410, investment name field 415 and investment data in a financial database according to one embodiment of the pointer 420. Investment ID field 410 stores a unique 32-bit present invention. Each financial data record 605 includes value corresponding to a particular investment. Investment investment ID field 610, and one or more return objects name field 415 stores a 16-byte character array of a name of 625(1)–625(N). Investment ID field 610 stores a 32-bit an investment. Investment data pointer 420 Stores a pointer 55 integer value uniquely identifying a financial record. Return to a linked list of financial data records related to an objects 625 (as described in FIG. 6b) store actual data values investment, which are stored in financial database 150d of returns associated with the investment represented by (described in more detail below). investment ID field 610. FIG. 5a depicts the structure of an alert rule record that is FIG. 6b depicts a data Structure for Storing a return object stored in an alert rules database 150c at a financial infor 60 according to one embodiment of the present invention. Each mation site 119 according to one embodiment of the present return object 625 includes a date field 630 and a value field invention. According to one embodiment of the present 635. Date field 630 stores a data object corresponding to the invention, each alert rule Specifies a percentile constraint of data of a return and value field 635 stores the value (dollar a resampled distribution for which a client 105 desires amount or otherwise) of the investment on the date Stored in notification. Clients 105 of financial information site 119 65 date field 630. may desire to be notified if the occurrence of a current event FIG. 7 depicts a data Structure for Storing data in a is extremely unlikely. AS described in detail below, financial function database 150f at a financial information site 119 US 6,349,291 B1 9 10 according to one embodiment of the present invention. FIG. 9a (reprinted from Efron and Tibshirani) depicts the Function database 150f Stores various function prototypes underlying theory of the bootstrap method. Ideally Statistical for functions to be performed on investment data, which are inferences are based on a known probability distribution F. used in performing resampled Statistical analysis of financial A parameter is a function of a known probability distribution data. For example, according to one embodiment function F. database 150f Stores function prototypes for gross rate of return, maximum drawdown and/or a monitor function. Each function record 705 includes function prototype ID Furthermore, generally financial data may not be modeled 710 and function prototype object 715. Function prototype parametrically because it is heavy tailed (i.e., non-Gaussian) ID field 710 stores a unique 32-bit integer value pertaining 10 and therefore, F is not known or ascertainable. For example, to a function prototype, which is used for identification with respect to financial data, an investor may desire to Study purposes. Function prototype object field 715 stores a 1024 a specific parameter of a financial investment Such as the byte character array of a function prototype. The Syntax for grOSS rate of return of a Stock over a certain period of time representing a function prototype Stored in function proto that is dependent upon knowledge of the true probability type object field 715 is variable. Practitioners skilled in the 15 distribution function for the investment. However, generally art will recognize that many data structures and techniques neither the PDF for an investment, nor the PDF for a may be utilized to represent function prototypes. According parameter Such as the groSS rate of return over a specified to one embodiment of the present invention, a maximum time period for the investment (which is dependent upon the drawdown function prototype is Stored in function database underlying PDF) is known. 150 based upon the following equation: For a set of returns 20 Resampled Statistical methods Such as the bootstrap (r-r): attempt to estimate the PDF of an unknown distribution Max. Drawdown = 1 +r (1 + r.) (1 + 2) (1 + r )(1 + 2)(1 + 3) (1 + r )(1 + 2)(1 + 3) ... (1 + r.) 1 1 1 1 (1 + r.) (1 + 2) (1 + r )(1 + 2)(1 + 3) (1 + r )(1 + 2)(1 + 3) ... (1 + r.) (1 + r.) (1 + r.) (1 + r.) 1 - Mi O O (1 + r )(1 + 2)(1 + 3) (1 + r )(1 + 2)(1 + 3) ... (1 + r.) (1 -- r1)(1 + r2) (1 + r )(1 + 2) O O O (1 + r )(1 + 2)(1 + 3) ... (1 + r.) (1 + r.) (1 + 2)(1 + ra) ... (1 + r. 1) According to one embodiment of the present invention, a using Sampled data. Typically, Sample data is available for grOSS rate of returns function prototype is Stored in function an investment that is dependent upon an unknown PDF F. database 150f based upon the follow equation: For a set of 40 F x=(x1, x2, ... x.) returns (r-rn): The empirical distribution function F is defined to be the discrete distribution that puts probability 1/n on each value Gross Rate of Return = || (1 + r.) - 1 X, i=1,2,... n. Fassigns to a set A in the sample space of X its empirical probability Prob{A}=#{xeA}/n, In, the pro 45 portion of the observed Sample X=(x1, . . . X). According to one embodiment of the present invention, a The plug-in principle is a method of estimating param monitor function calculates a number of 'up' or down days eters from samples. The plug-in estimate of a parameter for a given investment over a certain period. Thus, the 0=t(F) is defined to be 0=t(F) 910. These statistics are following equation describes a monitor function: referred to as Summary Statistics, estimates or estimators. 50 Resampled statistical methods attempt to determine the distribution of 0, an estimator of 0, derived from a sample X. Bootstrap methods depend on the notion of a bootstrap sample. If F is an empirical distribution with probability of Thus, for example, according to one embodiment, the 55 1/n for each of the observed valueS X, i=1,2,. . . ... Il, a maximum drawdown function, the groSS rate of return bootstrap Sample is defined to be a random Sample of size n function (as described above) and the monitor function are drawn from F, X*=(X,X2,...,x), where F->(x, *,x*, coded according to a predefined Syntax and Stored as a ..,x). The Star notion indicates that X* is not the actual function prototype in function database 150f data Set X, but rather a randomized, or resampled version of FIG. 8 depicts a data structure for storing plot information 60 X. The bootstrap data points X1,X2,.. .,X, are a random in a plot database at a financial information site according to Sample of size n drawn with replacement from the popula one embodiment of the present invention. Each plot type tion of n objects (x1, x2, . . . , X). Corresponding to a record 805 stores plot type ID field 810 and one or more plot bootstrap data set X* is a bootstrap replication of 6, 6*-s(x*) parameter fields 825(1)–825(N). Plot ID field stores a 915. The quantity S(x) is the result of applying the same unique 32-bit integer identifying a particular plot type. Plot 65 function S() to X as was applied to X (i.e., the statistical parameter fields 825(1)–825(N) store various parameters function of interest). For example, S() may be the gross rate relating to formatting of plots. of return of an investment over a Specific period of time. US 6,349,291 B1 11 12 Thus the bootstrap attempts to estimate a parameter of used in the resampling process. The plot type parameter interest 0=t(F) from an unknown distribution F using a Specifies a plot type identifier (i.e., 810) pertaining to random Sample X=(X1,X2,...,x). Given a random Sample formatting parameters to be used in generating a plot of X=(X,X2,...,x) and a statistic 0=S(X, F) that depends on the distribution results. sample and possibly the underlying distribution F, the dis In step 1015, back end server 140b requests the initiation tribution of 0, of a resampled statistical analysis process at RSAE 139. In particular, according to one embodiment of the present t(0=S(x,F)) invention, back end server 140b transmits the following vector to control server 140c at RSAE 139: is estimated by that of request resampling process (investment, function, peri (6*=S(x*F) ods (Q), bias, Sample size, replications, plot type) Back end server 140b then waits for completion of the FIG. 9b depicts a process for performing a bootstrap resampled Statistical analysis task. In Step 1020, back end method (a resampled Statistical method) to generate a dis server 140b determines whether RSAE 139 has completed tribution of bootstrap replications according to one embodi 15 the resampling process. According to one embodiment, upon ment of the present invention. In Step 920, a Sample Space completion of the resampling process, control Server 140c X is selected. In step 925, a statistical function based on the transmits a completion Signal to back end Server 140b and an Sample Space data is computed 6=t(F). In step 930, bootstrap address in output data area 160b where output data of a Samples X*=(X,x, . . . .X.), are generated from the resampled Statistical process is Stored. If the resampling Sample Space using a resampling process. In Step 935, a process is not completed (“no branch of step 1020), back bootstrap replication 6=S(x) is computed for each bootstrap end server 140b continues to wait for notification. If the sample using a desired function. In step 940, a plot of the resampling method is completed (yes' branch of step 1020), distribution of bootstrap replications (s(x*").s(x**) . . . in step 1025, back end processor 140b requests a graphics s(x)) is generated in order to estimate the distribution of plot from GRE 129. In particular, according to one 0. 25 embodiment, back end processor 140b transmits the follow FIG. 10 is a flowchart of steps for initializing a resampled ing vector to graphics rendering server 140e at GRE 149: Statistical analysis of financial data at a financial information plot(OutAddr, plot type, plot parameters). site 119 according to one embodiment of the present inven OutAddr Specifies an address in output data area, which tion. In step 1005, the process is initiated upon receipt of a Stores results of a resampled Statistical process previously request for a resampled Statistical analysis of financial data, conducted by RSAE 139, plot type specifies a plot type which is received via front end subsystem 129 (e.g., via an requested by a client 105 and plot parameterS Specifies HTML form). In step 1010, input parameters relating to a additional plotting parameters that may be required by GRE reSampled Statistical analysis are received Via client/gateway 129. Back end server 140b then waits for completion of the server 14.0a and transmitted to back end server 140b. plot. In step 1027, back end processor 140b determines According to one embodiment, the following parameters are 35 whether graphics rendering Server 140e has completed the Solicited from a client 105: investment; requested plot (i.e., whether graphics Server has transmitted function; a completion signal to back end processor). According to periods (Q); one embodiment, upon completion of a plot, graphics ren bias, dering Server 140e transmits a completion Signal to back end 40 processor 140b. Graphics rendering server 140e also trans Sample size; mits results of the plotting process in the form of plot data, replications, and which may be used to dynamically create an HTML page for plot type. transmission to a client 105. If the plot is not finished (no The investment parameter Specifies an identifier of an branch of step 1027), back end processor 140b continues to investment (i.e., 410) stored in investment database 150e. 45 wait for the completion Signal. If the plot has been com The function parameter Specifies a function prototype pleted (yes' branch of step 1027) in step 1029 and back end identifier (i.e., 710) stored in function database 150f. For processor 140b transmits the plot data results (e.g., HTML example, according to one embodiment, the function pro page) to client/gateway server 14.0a for transmission to totype may correspond to a function for maximum client 105. The process ends in step 1030. drawdown, grOSS rate of return or a monitor function as 50 FIG. 11 is a flowchart that depicts a set of preparation described above. The periods parameter Specifies a number steps performed by a control server 140c at a financial of periods for which a client 105 desires to evaluate an information site 119 to initialize a resampled statistical investment. For example, a client 105 may desire to perform analysis of financial data using a parallel processing. In Step a resampled Statistical analysis for the groSS rate of returns 1105, the process is initiated upon the receipt of a request of an investment over a 253-day period. The bias parameter 55 resampling proceSS vector from back end Server 140b as is a decimal value that is either -1 or between 0 and 1 that described above with reference to FIG. 10. Specifies the degree of randomneSS in the resampling pro In steps 1115-1119, control server 140c reserves appro ceSS. A value of -1 indicates that the resampling proceSS priate memory in Shared memory area 160a and output data should be conducted purely randomly. AS described in more area 160b and Stores appropriate Sample data for processing detail below, if the bias’ parameter is between 0 and 1, 60 in shared memory area 160a. In particular, in step 1115, a Sampling is performed So that b% of the Samples are up Sample Space is determined using the Sample Size param days and 1-b% of the samples are down days, where eter received in step 1105. Because financial database 150d b=bias. Thus, if b=1, only up days will be selected and if may store Samples for investments for many different time b=0 only down days are selected. The sample size periods, in step 1115, a set of relevant samples for the parameter Specifies a number of Samples to use in the 65 resampled statistical analysis requested by the client 105 is resampling process (the size of the X). The replications determined. In Step 1117, based upon the Sample size parameter Specifies a number of bootstrap Samples to be parameter, control Server 140c determines an amount of US 6,349,291 B1 13 14 memory required for Storage of Samples in shared memory factor (N) that minimizes the autocorrelation (a). The mini area 160a and allocates an appropriate memory block in mum lag factor (a) corresponds to the number of consecutive shared memory area 160a for Storage of the Samples. periods that are Selected at one time during the resampling Further, based upon the replications parameter, Server 140c proceSS. also determines an amount of memory to reserve in output In step 1135, the bias parameter received in step 1105 is data memory area 160b for storage of results of the resam analyzed. If no bias is Selected (i.e., bias=-1 and data is to pling process. In Step 1119, based upon the Sample size be selected randomly), control passes to step 1045 parameter, Server 140c retrieves financial data Samples from (“nobranch of step 1035). If bias<>0, in step 1040, a bias financial database 150d and stores these samples in shared initialization algorithm is performed as described in detail memory area 160a in the memory block previously reserved below. In step 1145, it is determined whether the sample in step 1117. In step 1120, process server 140c computes 0 Space data should be transformed. This determination is from the sample data stored in shared memory area 160a. In based upon the precise function requested by the client 105 particular, a Statistical function Such as the mean, median or (i.e., specified by function parameter). For example, if the Standard distribution is calculated by dividing the Sample Space into appropriate length periods. function is groSS rate of return over a specified period, no In steps 1125–1160, control server 140c executes a series 15 transformation Step is required. However, if for example, the of Steps to format and prepare the data for processing. function type is the monitor type, the Sample data is trans Specifically, in Step 1125, autocorrelation of the Sample formed to represent the sign of the returns only (i.e., -1 and Space data Stored in shared memory area 160a is analyzed. +1). Other variations will exist depending upon the type of Specifically, control Server 140c executes a process to cal functions implemented. If no transformation is necessary culate the autocorrelation and partial autocorrelation func (no branch of step 1145), control is transferred to step tions on the data stored in shared memory area 160a for a 1160. Otherwise (yes’ branch of step 1045) in step 1150, the number of different lag periods (a) and Stores the results in data is transformed and restored in Shared memory area temporary Storage. According to one embodiment, the fol 160b in step 1150. lowing equations are utilized to calculate the autocorrelation In step 1160, the variable M=Int(Q/N) is determined. The and partial autocorrelation functions for the data Stored in 25 variable M specifies the number of samples to select for shared memory area 160a. each resampling. In step 1165, server 140c executes a Wa