145
- Who provides the interface?
- How timely will the data be? When will it arrive?
No data source can do everything, but the trading/investment firm that waits for deliv-
ery of data only to learn later what they have bought is wrong deserves what they get.
Given competing vendors, we recommend buying from the vendor whose data integrates
best with your architecture regardless of cost.A trading firm refused to pay a data vendor for historical data. It was too expensive. Instead they used it
without permission in violation of their contract. They then complained about the fact the data was not
properly split adjusted. Their historical volatility was regularly wrong and caused bad trades. The money
lost on bad trades would have paid for the data. However, the CIO would not pay for data; in his mind it
should be free and clean. Eventually, the traders complained and forced the CIO out. Now the firm pays
for clean data.
With respect to price data, we recommend you pick a data vendor based on the fol-
lowing questions:● How do you calculate what you believe the closing bid/ask is?
● Can I pick the exchange?
● How do you handle uneven volume at the close?
● Can I get an off-close final price that I can use with my options?
● Can I synchronize my option closing price with my stock closing price?
● Do you use CUSIP or your own unique identifier?
● How do you handle mergers and spin-offs?We recommend that ideally you get the price at the time you want, with a unique identifier
that is consistent between the stock, the underlying of options, and the underlying of the
corporation ’ s bonds; ones that do not change with expirations, mergers, or spin-offs. The
nonofficial closing prices can be considerably cleaner since block trades and end-of-day
anomalies are not in the data. In currencies, there are multiple closes so traders can select
a time they believe gives the cleanest price. Why cannot traders in other markets (that are
now becoming 24 hour markets) select times?
The best data is the data where you understand its errors and issues, and works prop-
erly with your other software systems. For example, an algorithmic risk engine may be
integrated with data from a third-party provider using unique IDs. It would be foolish not
to buy data from the vendor since it is already integrated into your risk management and
backtesting engine.
When you look for fundamental data, we recommend buying it all from the same data
provider to ensure calculations have been consistently applied.We know a data vendor that offers their data through four different wholesale products. Believe it or not,
the values that are in the data may or may not be the same for the four. How do you know which numbers
are correct? The vendor claims that it comes from the same source, yet we can show where the data is
different. Legacy code must be making changes behind the scenes. To further complicate the problem, the
wholesale products have their own cleaning algorithms that further change the data. The scariest part is
that the vendor cannot solve the problem!
14.3. STEP 1, LOOP 2: PURCHASE DATA