Optimal OO Design for a Mixed Instrument Portfolio

Question

I'm designing a system designed to calculate various metrics on a set of holdings based on an initial day's holding and a feed of new trades and price ticks. The initial information comes from a WCF service with events for new trades and price updates.

The metrics required include market value (MV) that also needs to be aggregated into a hierarchy in various ways.

On the initial load we get a set of objects which look like:

Position [ Qty:double, Security:Security, Account:Account, Strategy:Strategy ]

Security [ Price:Money, Country:Country, Currency:CCY ]

Account [ Name:string, Fund:Fund, Manager:Manager, Currency:CCY ]

Strategy [Name:string]

Fund [Name:string]

Manager [Name:string]

The multiplicity looks like this:

• Manager (1) -manages-> (1..*) Account
• Fund (1) -is made up of-> (1..*) Account
• Account (1) -contains-> (1..*) Position
• Security (1) -can be in-> (1..*) Position
• Strategy (1) -contains-> (1..*) Position
• Strategy (1) -can exist in-> (1..*) Account

There are also a number of things that will happen during the day:

• Security price change - price of security changes from a -> b
• New position - There has been a trade
• Remove position - A trade has been cancelled

From the front end - we need to be able to view the data on a Manager/Fund level i.e.

• Manager 1
  • Total MV = $4800
  • Biggest country exposure = 43% UK
  • Total % MV = 100%
  • Components : List>
  • Countries : List>
• Manager 2
  • Total MV = $1200
  • Biggest country exposure = 90% USA
  • Total % MV = 100%
  • Components : List>
  • Countries : List>
• Fund2
  • Total MV = $1200
  • Biggest country exposure = 90% USA
  • Total % MV = 100%
  • Components : List>
  • Countries : List>

Or at a security level

• Sec 1
  • MV = $1000
  • % of Manager 1 = 67%
  • % of Fund 1 = 48%
• Sec 2
  • MV = $2000
  • % of Manager 3 = 12%
  • % of Fund 2 = 4% ...

Entities will be loading in an identity map to ensure they exist only once in the system and so a change to a security's price will be reflected in the relevant positions.

I've look at various way's to represent this, conceptually event sourcing (http://martinfowler.com/eaaDev/EventSourcing.html) looked promising although we only need to look at the current state so is probably overkill.

My current line of thinking is to represent Manager, Fund and Account as Portfolio object:

And to use the composite pattern (Portfolio : Component, Position : Component) + visitor patterns to encapsulate hierarchy + calculation logic respectively.

Portfolio [ object: Represents, Components: ListOfComponents, Accept(Visitor v) : state ]

Component also has MV (in $) (these can be calculated using a visitor)

Let MV [position] = *Qty * Security.Price*

Let MV [portfolio] = sum of all MV of positions below.

So we have the following composite structure:

• Portfolio (Represents: Fund A, MV: 500)
  • Portfolio (Represents: Account A, MV: 500)
    • Position1 (MV: 300)
    • Position2 (MV: 200)
• Portfolio (Represents: Fund B, MV 1600)
  • Portfolio (Represents: Account B, MV: 1000)
    • Position3 (MV: 800)
    • Position4 (MV: 200)
  • Portfolio (Represents: Account C, MV: 600)
    • Position5 (MV: 200)
    • Position6 (MV: 400)

My first problem is with the representation of various levels of percentage - the system needs to be able to represent what percentage of Fund B and Account C does Position 5 make up?

Position 5 MV / Account C MV = 200 / 600 = 33%

Position 5 MV / Fund B MV = 200 / 1600 = 12.5%

The values can be calculated using the visitor but where do we store the results of the calculation?

In a dictionary in the Position 5 like so: Dictionary percentMV = {[Fund B, 12.5%], [Account C, 33%]}

Or in the relevant portfolios in which case, all of the individual positions would need to be referenced by Fund B? We would also need Position 5 in terms of the Manager that manages it.

The other concept I'm having difficulty with is country exposure metric - each security represents a certain amount of risk associated with a country. So with a portfolio of two positions: - MV of $100, Security Country = UK - MV of $300, Country = USA

The portfolio has a Country exposure of - 100/400 =25% UK, and 300/400 = 75% USA.

What's the best way to generalize this concept into the design i.e. where's the best place store the data?

Answer 1

The Portfolio is a good idea to represent arbitrary segmentations of a population of positions; be sure to also take a look at the Specification pattern (pdf) for a generic way to specify the inclusion criteria for the portfolios. In fact, the portfolio pattern described in Analysis Patterns (Martin Fowler) is compared to the Specification pattern described in Domain Driven Design (Eric Evans).

Now, a Portfolio just represents an inclusion criteria of positions. To actually aggregate the positions in different ways, take a look at the Enterprise Segment pattern (described in Analysis Patterns). An Enterprise Segment is a slice of the different dimension elements that comprise your domain (in your case: security, manager and fund) on which interesting measurements can be aggregated, calculated and stored (like market value and country exposure).

Also, be sure to take a look at more fundamental patterns like Money, Quantity and Range; and always use decimals or integers to store monetary values or rates for monetary calculations.