6.2.3 Data Sources
There are many sources of financial data suitable for value-at-risk analyses. The most common include:
- exchanges,
- broker or dealer quotes,
- data vendors,
- real-time data feeds, and
- trade tickets.
All exchanges record detailed information on transactions. This is used for various purposes, including:
- audit,
- dispute resolution,
- distribution of real-time price information via data feeds, and
- determination of settlement prices.
Not all financial data is available publicly, but most exchanges distribute price histories indicating, among other things, high, low, closing, and settlement prices for each trading day.
Indicative prices may be obtained directly from brokers or dealers. It is best to standardize this process so the operational definition of the time series remains stable over time. Quotes may be obtained from one or several parties. Be candid with quoting parties. Let them know that the quotes are for risk management purposes only, and seek their commitment to provide the quotes on an ongoing basis. Many brokers or dealers are happy to provide this service for institutions with whom they have a profitable business relationship.
Data vendors collect data from multiple sources, preprocess it, and distribute it as time series. Costs vary but are generally modest compared to the cost of constructing and maintaining one’s own time series.
It is important to understand what sources data vendors use for data as well as what preprocessing they perform. If possible, it is best to purchase data prepared specifically for risk analyses. Some data vendors distribute covariance matrices for key factors instead of time series. If possible, avoid using these. As we shall discuss in Chapter 7, different analyses may yield very different covariance matrices from a given time series. The choice of technique is a significant design decision for avalue-at-risk model that should not be left to a data vendor.
Most trading organizations purchase real-time data feeds for their traders. Values can be captured from these periodically—perhaps at an appointed time each day—and used to construct time series. Data collected in this manner may require extensive cleaning. Because the data is delivered in real time, quality controls are minimal. Even in liquid markets, data feeds may lag the markets. Also, most vendors place restrictions upon use of their real-time data, and these may prohibit the accumulation of historical databases.
If an institution is active in a particular market, it may capture transaction prices directly from its own trade tickets, or subsequently from its trade accounting system. This should be considered only as a last resort if alternative data sources are unavailable. It is feasible only for market participants who trade in volume—primarily market makers. Cleaning the data is expensive and labor intensive. There is the risk of losing the data source if, for some reason, the institution is forced to stop trading in a particular market for a period.