System Dynamics Modeling and Simulation
Transcription
System Dynamics Modeling and Simulation
System Dynamics Modeling and Simulation An Overview of Booz Allen Hamilton’s System Dynamics Modeling Capability by Chip Jansen [email protected] 2 1. Overview of System Dynamics System Dynamics (SD) is an advanced simulation methodology that helps improve understanding of complex system and causal structures by focusing on system behavior over time and system behavior in response to stimuli. Originally developed in the United States at the Massachusetts Institute of Technology’s Sloan School of Business in the 1950s and 1960s, SD applies the mathematics of feedback control to business or mission systems. SD helps link policies to their impacts by simulating system responses to policy changes. This furthers knowledge about the causal mechanisms within a complicated system, allowing for better policy decisions. SD models can be rapidly built and modified to increase understanding of the simulated system. The primary language of SD is stocks and flows. Stocks are accumulations, such as stores of cash, inventory levels, or quality-adjusted life years. Flows are simply the rates at which those stocks accumulate or deplete, and can include a variety of inputs to their calculation. One aspect of complicated systems that SD handles more gracefully than some other methodologies is the notion of system feedback, which describes how even seemingly simple systems display baffling nonlinearity. A sample model architecture is illustrated in Exhibit 1. Exhibit 1 | System Dynamics Modeling Number of Parts (Parts) Current [+OT] Casual Tracing - Backlog Queue (System, Maintenance) Current maintenance complete S Experience Level Through Put or Availability Issue S Experience Loss $$ Backlog Queue (System 1, Overhaul) 8 actual final inspection time reorder 4 percent rework required 0 Backlog Queue (System 1, Repair) 8 actual initial inspection time 4 percent pass initial inspection Backlog Queue tagged for shop work 0 Backlog Queue (System 1, Modification) 8 tagged for inspection 4 Training S Attrition (Backlog Queue) 0 to shop shop work constraints Backlog Queue (System 2, Overhaul) 8 4 Uses of current variable S Training $$ Variable Definition Select a new variable to trace Change subscript selection Return to Analysis Control Backlog Queue (System 2, Repair) 8 8 4 4 0 Help? 0 Backlog Queue 0 Backlog Queue (System 2, Modification) 8 1 2 1 2 3 4 5 6 3 4 5 6 7 8 9 10 4 0 0 2.5 5 7.5 10 YEARS Source: Booz Allen Hamilton 2. Booz Allen Hamilton’s Approach to System Dynamics Modeling and Simulation Booz Allen Hamilton, a leading strategy and technology consulting firm, uses a systematic approach to developing our SD models. This approach revolves around a combination of qualitative, quantitative, and systems thinking analysis, including appropriate combinations of statistical analysis techniques and data mining. Systems thinking focuses on the whole view of a complex system, enabling analysts, for example, to identify non-intuitive behavior or results. Systems thinking has its foundations in SD. Our SD model design process is depicted in Exhibit 2 on the next page. This approach provides a powerful method for blending qualitative and quantitative data, as well as systems thinking techniques together with traditional analysis techniques, to better understand and model the complexities of systems across a variety of domains. 3 2.1 Define Goals, Objectives, and Desired End-State Exhibit 2 | System Dynamics Modeling Approach 2 SELECTED MODEL THEMES IDE Risk Mitigation Training & Education Personnel Management Acquisition Decisions Organizational Re-structuring Technology/Capability Assessment Incident Management Strategy Formulation Informative Surveys Tactical Response Collaboration 3 G L O O E NE R A TE PS & IN CASUAL ITIA L D ESIG N REVI EW & R OUT EPO PU RT T 7 E GOAL, OBJEC DEFIN RED END TIVES -STATE & DESI NS, PTIO UM S ASS AINT IFY STR NT CON & 2.2 Identify Assumptions and Constraints 1 SIS NALY D E S IG N A X E C U T E E PLAN & Booz Allen Hamilton works with clients to define principal goals and strategic priorities that drive towards a desired end state. In addition, both Booz Allen Hamilton and clients will work together in defining mission success indicators against core operational requirements, identification and prioritization of risks, validation of mission priorities, and identification of possible stress test scenarios. 6 4 Critical to any modeling and simulation effort is a T RU C REF TE ST & T S clear understanding of the assumptions that drive N IN E C O D EL MOD MO EL 5 the model and the constraints under which it will operate. Working with clients and key stakeholders, Source: Booz Allen Hamilton we develop a candidate set of assumptions and constraints that will govern the model’s behavior. This candidate list is then subjected to a detailed review against the requirement, goal, or objective it impacts to ensure it is in line with mission priorities and operational requirements. 2.3 Cause-and-Effect Diagramming Starting with system relationships, Booz Allen Hamilton utilizes facilitated sessions with subject matter experts and relevant stakeholders, to develop relationship maps between system entities using Causal Loop Diagrams (CLDs), as illustrated in Exhibit 3. From these CLDs we can then begin defining the primary system variables and critical interactions. Primary system variables serve as “levers of change,” which allow stakeholders to vary the conditions under which the system operates in order to observe its response. Critical interactions define specific interactions among all the different system components and serve as a basis for completing multi-factor scenario analysis of the identified primary system variables. In addition, critical interactions allow for the definition of feedback loops and for the examination of multi-order effects. Exhibit 3 | Causal Loop Diagram + – = Change in SAME Direction = Change in OPPOSITE Direction A B = Delayed Effect & = Change in EITHER Direction Possible Implement Stop/Loss Pressure to Implement Stop/Loss Sea Tour Length Resignations Shore Tour Length Readiness Accessions Source: Booz Allen Hamilton 4 Training Throughput Training Turbulence Mandated Requirements 2.4 Data Collection and Model Construction Utilizing data mining, authoritative data sources, and other data gathering techniques, we gather and analyze information supporting the primary elements of the system. Logical relationships defined in 2.3 are broken down into further system sub-components, inputs and outputs are finalized, and business rules and quantitative equations that regulate system behavior are defined. Next, a quantitative model is constructed that will serve as the platform for performing multi-factor scenario analysis and stress testing the impacts to the overall system. While typically these models have been built using Ventana System’s Vensim application, Booz Allen Hamilton has also constructed SD models in isee systems’ iThink and the AnyLogic Company’s AnyLogic. 2.5 Test and Refine Model Testing ensures that all functional requirements, as specified in 2.1, have been met and verifies the accuracy of the data employed and the operation of the model. Clients, key stakeholders, and users play an important role in this step and often serve as the final validation check of the model. Any changes, discrepancies, or additional refinements captured during this phase are quickly addressed and incorporated into the final version of the model. 2.6 Design Analysis Plans and Execution System shocks, what-if analysis, tradeoff evaluation, decision making, and scenario analysis are just a few of the varied outputs of this phase in the process. Working in conjunction with clients and key stakeholders, Booz Allen Hamilton brings together the strategic factors, identified in 2.1 and 2.2, and combines them with the quantitative processes of the model, developed during phases 2.3 to 2.5, to develop and execute analysis plans against organization goals. Utilizing the developed model, we draw out system/organizational characteristics, positive and negative correlations, risk/non-risk factors, and trade-offs affecting the system/ organization. 2.7 Review Output and Report Results are then examined and summarized, recommendations are collected, and detailed analysis of the model’s results are captured and presented to both clients and key stakeholders. This includes an overview of analysis plans executed, key findings from the model and any identified future issues that an organization may need to consider. In addition, as illustrated in Exhibit 2, this does not necessarily have to be the end of the overall process. Insights gained through the analysis of model data can be injected back into previous process stages for further refinement of objectives, goals, assumptions, constraints, data, and even the model itself. As such, clients and stakeholders can go through this process multiple times, further refining the end product or products. 2.8 Overcoming Potential Risks There are many risks associated with the creation of any model. Risks such as diving too deep into the system and not looking at the system as a whole, or misinterpreting model outputs, by focusing on single numerical outputs rather than looking at overall trends and behavior, are common in any systems thinking or SD development process. Booz Allen Hamilton mitigates these risks by ensuring at every stage of the project that both client and key stakeholders understand the intended goal, objectives, and desired end state of the project. Our investigative processes are rooted in sufficient analytical rigor to enable complete traceability and defensibility. We coordinate with stakeholders early and throughout the process to ensure their buy-in and receive feedback, avoiding the expenditure of time and effort developing items that cannot be successfully or effectively executed. 5 3. Past Client Engagements 3.1 Diabetes Simulation in the United Arab Emirates (UAE) Program Description Booz Allen Hamilton conducted a proof of concept for various firms and government agencies in the UAE to demonstrate the debilitating impact of diabetes on an individual company. Technical Approach and Technical Resources Used We conducted a model-supported game, working on the game design and building an SD model using Vensim, as illustrated in Exhibit 4. We developed and used a seminar game design that included capture templates for player outputs. These player outputs were then fed into the model to provide a deterministic result of the player decisions during each move. Program Outcome, Significant Accomplishments, and Impact The game demonstrated to UAE business leaders the impact of diabetes on the workforce of a single company, both from financial and human perspectives. Exhibit 4 | UAE Diabetes Simulation Model <UnXPD to NDE> <UnDXTTH Out> <UnXTTO Out> <NEH Out> New Employees <NOPD Out> UnDXPDD <OPD Out> <UnDXPD to NDE> <UDXPDDH Out> <UnDXTTCH Out> <UnDXTT to DXTT> UnDXTTD UnDXTTCD <UnDXTTCO Out> NDE to UnDXPD UnDXPD to UnDXTT UnDxTT to UnDxTTC UnDX PreDiabetes UnDXPD to NDE <Employee Initial Population> Non Diabetic Employees UnDX Type Two UnDX Type Two Complicated <UnDXPDD> UnDxPD to DXPD <UDPDR Out> <DXTTO Out> UnDXTT to DXTT <TTSD Out> UnDXTTC to DXTTC <TTCSD Out> <PDSD Out> Initial DX PD Initial DX TT DX PreDiabetes DXPD to NDE <ENDDH Out> <NDE to UnDXPD> Source: Booz Allen Hamilton 6 NDE Deaths DXPDD <DPDR Out> UnDX Type Two DXTT to DXTTC DXPD to UnDXTT <DXPDDH Out> <DXPD to NDE> DXTTD <DXTTH Out> <DXTTCO Out> UnDX Type Two Complicated DXTTCD <DXTTCH Out> 3.2 Financial System Modeling – Central Bank of Azerbaijan Program Description The Central Bank of Azerbaijan (CBA) contracted with Booz Allen Hamilton for the purpose of understanding the effects of policies and law on financial transactions and the behavior of citizens when conducting financial transactions, while at the same time understanding risk to the financial system. Technical Approach and Technical Resources Used We designed and implemented an SD model of CBA’s financial network using the Vensim application, as depicted in Exhibit 5. Program Outcome, Significant Accomplishments, and Impact Results of the modeling effort showed CBA where improvements in efficiency could be made and where it should focus on keeping customers interested in bank programs. Those programs, in turn, need to be carefully managed in order to ensure the stability of the bank in question. Exhibit 5 | CBA Financial System Model Domestic Loans in Good Standing to Past Due Process Flow <Domestic LL to GSL> <PD NToG> <Domestic UL to GSL> Domestic PDL to GSL Pulse <Domestic DL to GSL> Loans to Performing <Domestic New Loan Demand> <Domestic GSL to PDL> <Domestic Lending Funds> <Domestic PDL to UL> <Domestic PDL to GSL Warmup Amount> <Past Due Rate> New Domestic GSL Clear Domestic PDL Initial Domestic GSL Domestic GSL Sold Domestic GSL to PDL Initial Domestic PDL <Domestic Principle Payments> Domestic PDL to UL Time <Domestic Principle Interval Payment> Domestic Unsatisfactory Loans Domestic PDL to UL Domestic PDL to UL Pulse Domestic GSL Repayment Window <Domestic GSL Sold> Initial Domestic UL Domestic GSL Interest Paid Total Domestic GSL Payments <Domestic Good Standing Loans> <Bank Has Failed> Doubtful Loans Domestic UL to DL Time <UN NToG> <Bank Has Failed> <Domestic GSL Sold> Domestic UL to DL Domestic DL to LL Time Domestic UL to DL Warmup Amount Domestic DL to LL Pulse Initial IBL Clear Domestic DL <PD NToG> <Time> <Domestic GSL to PDL> <Domestic PDL to UL> <Domestic UL to DL Time> Domestic DL to LL Domestic UL to LL Warmup Amount <DO NToG> <Time> <Domestic DL to GSL> <Loan Default Rate> <Domestic UL to DL> <Time> <Domestic PDL to UL Time> Domestic PDL to GSL Warmup Amount Inter Bank Loans <Domestic DL to GSL Warmup Amount> <Clear Bank Values> Initial Domestic DL Domestic UL to GSL Warmup Amount Domestic DL to GSL Warmup Amount <Domestic DL to LL Times> Interbank Lending Process Flow Bank Failure IBL Domestic DL to GSL <Bank Has Failed> Domestic UL to DL Pulse Domestic PDL to UL Warmup Amount <Domestic GSL Interest Rate> <Bank Has Failed> <Domestic UL to LL Time> <Domestic UL to GSL Warmup Amount> <Domestic UL to GSL> <Domestic PDL to GSL> Domestic Past Due Loans Domestic GSL Principle Payments Clear Domestic GSL <Domestic UL to DL Time> <Bank Has Failed> Daily Past Due Rate Domestic Good Standing Loans Domestic UL to GSL Clear Domestic UL <Domestic UL to DL> Domestic DL to GSL Pulse <DO NToG> <UN NToG> Domestic UL to GSL Pulse <Clear Bank Values> Domestic PDL to GSL <Bank has Become Non Bank Entity> <Bank Has Failed> <Clear Bank Values> Domestic Non Performing and Default Process Flow <Domestic PDL to UL Time> Domestic Loss Loans <Domestic Good Standing Loans> DGSL Minus Provision <GSL Provision> <Domestic Past Due Loans> DPDL Minus Provision <PDL Provision> <Domestic Unsatisfactory Loans> DUL Minus Provision <UL Provision> <Doubtful Loans> DDL Minus Provision <DL Provision> <Domestic Loss Loans> DLL Minus Provision <LL Provision> <Bank Has Failed> Daily Default Rate Clear Domestic LL <Clear Bank Values> Domestic Defaults Initial Domestic LL Domestic LL to GSL <Bank Has Failed> Daily Non Performing to Performing Rate Source: Booz Allen Hamilton 7 3.3 Armenia Tax Improvement Program – USAID/Armenia Program Description Booz Allen Hamilton developed a simulation model to assist the United States Agency for International Development (USAID) and the Government of Armenia in investigating ways to increase tax revenue collected as a percentage of gross domestic product (GDP) through more effective, equitable, and transparent tax administration. A pilot regional inspectorate was identified to test new procedures, policies, and systems prior to roll out at a national level. A Modeling & Simulation software program and workshop were used to evaluate the impact from a range of potential changes. Technical Approach and Technical Resources Used We designed and implemented an SD model of Armenia’s Tax Service using the Vensim application, as illustrated in Exhibit 6. Program Outcome, Significant Accomplishments, and Impact Expected results include increased collection of tax revenues as a percentage of GDP with a broader tax base, reduced reliance on value-added tax, improved administration of direct taxation, and enhanced information management systems. The study also showed achievement of these objectives will also improve the overall transparency and consistency of tax collection. Exhibit 6 | Armenia Tax Service Model <Time> Annual Requested Audits Monthly Timer Supervision Management Capture Possible Audits Additional Info Process Time Total Requests per Month Requests In Use Variable Audit Workload Requests Out Total Requests Inspectorate Staff Requests for Information Information Generation Generate Requests Send Requests <Final Report> Liquidation Request Rate Budget Relationship Requests DP Information Production Delay Information Provided Goods Registration Requests Other Requests Audit Requests Impact of S and P adoption on process rates Annual Requested Audits Final Information Collected Receive All Requests Impact of S and P adoption on Info Rates Average Audit Workload Sys and Proc Improvements Head of TI Info Assignment Delay Audits to be Considered Consider for Audit Use Capacity Audit Workload Impact of S and P adoption on Productivity <Audit Capacity> Green parameters are user inputs, Red are graphical lookups, Gold are used for debugging logic. Set Constant Defaults Source: Booz Allen Hamilton 8 Next Previous Copy to Clipboard Impact of S and P adoption on Staff Use <Send Request> Print <Shared Assess Switch> Return 3.4 Measuring Shocks to Ethiopia’s Agriculture System Program Description Booz Allen Hamilton conducted a proof of concept study to enable more robust analyses to determine the multiple impact of inputs (e.g., climate, conflict, geography, and natural resources) on a wide range of micro-indicators (e.g., farmer livelihood) and macro-indicators (e.g., GDP, exports, and employment). Technical Approach and Technical Resources Used As depicted in Exhibit 7, we designed and implemented an SD model, using the Vensim application, of the agricultural system in Ethiopia to test shocks to commodities and the agriculture sector by shifting “levers” such as climate and conflict, to identify vulnerabilities in areas such as food security and trade. Program Outcome, Significant Accomplishments, and Impact The simulation demonstrated to public policy makers the effects on individual livelihoods that various shocks on a country’s agricultural system can have, including changes in income and kilocalories available to be consumed. Exhibit 7 | Ethiopia Agriculture System Model Income Sources - NHB (Very Poor) 1.0 0.8 0.6 0.4 0.2 0.0 Yemen Sudan Djibouti Series Somalia 0 0 0 0 0 Cash Sub Crop Income Labor Income Other Cash Livestock income Vensim Model Directory C:Documents\020456\Desktop Vensim Model Name AAIP_v17.vpm Vensim Simulation End Time (Days) 12 Report Results Out By Month (30 Days) Simulation Start Date 01/01/2012 Simulation Run Name AAIPBaseline Run Simulation Food Sources - NHB (Very Poor) Current Date 10/17/2012 12/06/2014 09/07/2014 06/09/2014 03/11/2014 12/11/2013 09/12/2013 06/14/2013 03/16/2013 12/16/2012 09/17/2012 06/19/2012 03/21/2012 12/22/2011 WMB 09/23/2011 SWB 12 10 8 6 4 2 0 06/25/2011 Better Off 0 Other Food 03/27/2011 Middle SME 0 Purchased Food 12/27/2010 Poor NWE 0 Grown Food 09/28/2010 NMC 0 Food Aid 06/30/2010 Wealth Bands NHB Date NHB NMC NWE SME SWB WMB 0 01/01/10 100% 100% 100% 100% 100% 100% 395 01/31/11 10% 10% 10% 10% 10% 10% 696 11/28/11 10% 10% 10% 10% 10% 10% 1097 01/02/12 100% 100% 100% 100% 100% 100% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% Series 1 0% 0% 0% 0% 0% 0% Series 2 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% Household Consumption - NHB (Very Poor) Livelihood Zones Very Poor 0 Livestock Food Income 04/01/2010 Kenya Rainfall Modifier Sim Time Series 01/01/2010 Uganda 1.0 0.8 0.6 0.4 0.2 0.0 END Source: Booz Allen Hamilton 9 3.5 Financial System Modeling – 2008 US Financial Market Collapse Program Description Booz Allen Hamilton conducted a brief study in 2008 on ways to mitigate the impact to the economy of the ongoing US financial crisis. This study analyzed the impact on the overall financial system of key decisions the US government could make with regard to the Troubled Asset Relief Program (TARP). The study also looked at the counter-party exposure risk between the top 30 US banks and the effect of one bank’s failure on the rest of the banks in the system. Technical Approach and Technical Resources Used We designed and implemented an SD model of the US financial network using the Vensim application, as illustrated in Exhibit 8. Program Outcome, Significant Accomplishments, and Impact Results of the modeling effort showed the inherent dangers associated with counter-party risk among the top 30 banks to include system-wide failure due to a large bank collapse. In addition, the study also provided policy makers the means to gauge levels of effectiveness, with regard to TARP, given various system conditions and trends. Dollars Total Bank Failures 0 150 300 Total Bank Failures: Test Run 600 750 Loan Delinquency To Default Control Fed Lending Threshold 00 0 Derivative Scaler 00 0 0 Fed Ratio 00 0 AIG Losses On/Off 00 0 Fed On/Off 00 0 AIG Loss Time 00 Inter Bank Lending Controls Commercial Loan Default Percentage 00 0 Consumer Loan Default Percentage 00 0 Real Estate Loan Default Percentage 00 Receivership Controls 00 0 Receivership Return on Asset Sale 00 0 Consumer Loan Default Percentage 00 0 Time to Forget Financing Denied 00 0 Receivership Delinquency Modifier 00 0 Commercial Buyer Modifier 00 0 Real Estate Buyer Modifier 00 General Lending Controls 0 01 0 TARP Repayment Start Time 00 0 TARP Repayment Window 00 Source: Booz Allen Hamilton 0 Starting TARP Value 0 150 300 4 3 2 1 0 1.75M 0 Short Term Funding Availability 450 Time (Day) 600 750 900 750 900 750 900 Total Bank Failure Asset 0 150 300 Total Bank Failure Asset: Test Run 450 Time (Day) 600 Total Bank Failure Equity Inter Bank Lending Threshold TARP Repayment On/Off Total Bank Failures Liability Loan Asset Sale Controls 0 TARP Controls 10 900 4 3 2 1 0 Total Bank Failure Liability: Test Run Losses Controls FED Controls 0 450 Time (Day) Dollars 4 3 2 1 0 Dollars DMNL Exhibit 8 | 2008 US Financial Network Model 4 3 2 1 0 0 150 300 Total Bank Failure Equity: Test Run Commercial Loan Controls 00 0 0 Commercial Loans Delinquency % 450 600 Time (Day) Consumer Loan Controls 100 Commercial Loans Nondelinquency % 100 0 0 Consumer Loans Delinquency % Real Estate Loan Controls 100 Consumer Loans Nondelinquency % 100 0 0 Real Estate Loans Delinquency % 100 Real Estate Loans Nondelinquency % 100 11 International Office Locations Principal Headquarters Office Principal Offices OperatingOperating OfficeOffices Global Headquarters About Booz Allen Hamilton Booz Allen Hamilton has been at the forefront of strategy and technology consulting for nearly a century. Today, the firm provides services to US and international governments in defense, intelligence, and civil sectors, and to major corporations, institutions, and not-for-profit organizations. Booz Allen Hamilton offers clients deep functional knowledge spanning strategy and organization, engineering and operations, technology, and analytics—which it combines with specialized expertise in clients’ mission and domain areas to help solve their toughest problems. Booz Allen Hamilton is headquartered in McLean, Virginia, employs approximately 25,000 people, and had revenue of $5.86 billion for the 12 months ended March 31, 2012. To learn more, visit www.boozallen.com. (NYSE: BAH) For more information contact Daniel Whitehead Senior Associate [email protected] +971-50-442-8634 Chip Jansen Lead Associate [email protected] +971-2-691-3600 To learn more about the firm and to download digital versions of this article and other Booz Allen Hamilton publications, visit www.boozallen.com. www.boozallen.com/international ©2013 Booz Allen Hamilton Inc. 01.001.13