Modeling Market Impact in VWAP Execution Backtests

# What is an effective way of backtesting VWAP execution?


# What is an effective way of backtesting VWAP execution?












From Optimal Trading Strategies :

> There are two main reasons why traders execute orders using a VWAP trading strategy. First, a VWAP strategy is the trading strategy that minimizes market impact cost. Traders executing orders utilizing VWAP strategies are interested in participating with market volumes so to cause the least disruption to the supply-demand imbalance in each period. Second, for many traders their execution performance is measured against the VWAP benchmark. An average execution price more favorable than the VWAP price benchmark is considered high-quality performance while an average execution price less favorable than the VWAP benchmark is considered inadequate performance.

But, when it comes to backtesting, the market impact is thus assumed to be zero. One is assuming he does not affect the market as the VWAP benchmark price stays untouched. So how can one backtest his VWAP strategy effectively and more realistically?

## Answer by chrisaycock (score 6, accepted)

https://quant.stackexchange.com/a/1134

This isn't exclusive to VWAP; any assumed trade price (NBBO, Arrival Price, etc) has the same vulnerabilities. Many shops often lump market impact with slippage and transaction costs when modeling the difference between the ideal price and the realized price.

To model the impact during a backtest for a given trade price, assess a penalty:

- This penalty can be fixed (X basis points added to each buy price) or it can be variable (Y percentage extra cost for each buy price). One could use a combination of both.

- For variable penalties, the percentage can be a single number for all trading activity, or it can grow quadratically with respect to the size of the trade. An aggressive penalty models removing liquidity in the order book.

- To be really swanky, one could have a different penalty for each unique instrument in the portfolio.

Many shops consider their transaction-cost models to be very proprietary. It takes a lot of post-trade analysis to determine proper values for X and Y.

## Answer by Rich C (score 4)

https://quant.stackexchange.com/a/1138

To get a feel for it think about an extreme scenario. Suppose I have an order to buy $100bln VWAP in IBM over the course of one day. My "cost" relative to VWAP will be near zero, b/c I will be on the buy side of every trade that day. However, my market impact will be extremely high b/c IBM will fall like a rock the next day, leaving me with huge losses. This is your true cost.

The bottom line is trading will create impact - that is part of your cost. It is increasing in increasing trade size and decreasing in increasing liquidity of the asset. VWAP merely spreads the trade out to reduce the impact. Look at Almgren and Chriss for a discussion of transaction cost functions, or talk to the provider of your VWAP algo for more details.

## Answer by Meh (score 0)

https://quant.stackexchange.com/a/1140

A factor to take into account: there are algorithms searching for other VWAP algorithms present in the market and then gaming them.

http://www.qsg.com/PDFReader.aspx?PUBID=722