Given the fundamental nature of VWAP in the trading world, it's no surprise that there are algorithms or algos designed around it. Those VWAP execution algorithms have been developed to help traders achieve or outperform the VWAP benchmark for a particular order. The basic TWAP strategy divides the order into uniform segments executing consistently over the designated period. A VWAP algorithm, however, does not just demand regular execution, but also trading in precise proportions in line with fluctuating market volumes. These proportions depend on the realized volume over the entire relevant period, which unfortunately will only be known at the end of the period. So how can this issue be addressed? A common way is to use historical volume profiles, which are given by averages of trading volumes observed in the past. Based on this approach, the basic mechanism of a VWAP algorithm works as follows. First, we have to determine the number of shares we want to buy or sell. The algo then uses historical data to predict the volume distribution throughout the day. Based on the volume forecast, the algo slices the order into smaller chunks known as child orders to be executed at different times. These child orders will then be automatically released at predetermined times.

Let's look at an example. Let's say we're aiming to buy 10,000 shares at VWAP over a two hour time period from 9:15 AM. Let's assume that historically the security has traded in total volumes of 258,000 over the two hour period with an average of 43,760 units at 9:15. This corresponds to 16.96% of the total volume of the time period. So the algorithm will trade 1,696 shares at the price available at 9:15. The same methodology is applied to all other given points in time, and as a result, a total of 10,000 shares will have been purchased at the end of the two hour period.

However, if we calculate the average price realized by the basic VWAP algorithm using the determined child order sizes, we get a result that differs from the actual VWAP. More specifically, the algo achieved an average price of 81.7495, whereas the VWAP for the period was 81.7516. What is the reason for this difference? Remember, our basic VWAP algo predicted the volume distribution over the execution period using historical data and then statically executed the child orders. The realized volume distribution during the period, however, was different. As we can see on the slide. The realized volume percentage was lower than the historical one for the first part of the period. This is also when the share price was relatively low in the second half Of the period realized volumes were higher than the historical ones, and share prices were relatively high. So the algorithm based on the historical volume distribution bought more shares at lower prices and fewer shares at higher prices than realized by the actual VWAP, which leads to the lower average price of the algorithm compared to the VWAP. So in this example, the algorithm beat the benchmark, but this example also shows the main issue of the VWAP program. If the realized volume distribution differs meaningfully from the historical volume profile, the average price achieved by the algorithm might differ significantly from the benchmark. While this could result in out performance, it could of course lead to significant underperformance as well.