Bridgeton Orders Event Study

Nov 27, 2018 12 min read Bridgeton, Events Study, Market Positioning

1 Introduction

The aim is this study is to see if we can find repeatable and economically viable trading signals from the Bridgeton Order data. The Bridgeton order flow data is created using trend following and other systematic strategies that are well known in the literature. There are only a finite number of ways that you can generate entry and exit signals of trend following systems. They reverse engineer these trend strategies to find the price levels where they expect to see large players in the markets climb in and out of positions.

The table below shows an example of what the data looks like. Here we show the data for Chicago wheat on the first two days of january 2013. The YearMo column defines the active contract, in this case 201303, i.e. the March 2013 contract. BuySellidentifies if it is a buy or sell order. Order lists the price at which the order will be executed. VolumeEst is their estimated number of contracts that will be traded at the order price. Notice that the BuySell column contain Close values for both Order and VolumeEst. These correspond to the closing price and estimated net speculative postion (TFR NET).

date	YearMo	BuySell	Order	VolumeEst
2013-01-01	201303	Buy	855.80	2941.18
2013-01-01	201303	Buy	929.75	8823.54
2013-01-01	201303	Sell	759.26	17647.08
2013-01-01	201303	Buy	865.00	2941.18
2013-01-01	201303	Buy	912.78	2941.18
2013-01-01	201303	Buy	865.00	11764.72
2013-01-01	201303	Close	778.00	-8823.53
2013-01-02	201303	Buy	864.75	2941.18
2013-01-02	201303	Buy	912.78	2941.18
2013-01-02	201303	Buy	864.75	11764.72
2013-01-02	201303	Sold	759.27	17647.08
2013-01-02	201303	Buy	854.12	2941.18
2013-01-02	201303	Buy	929.75	8823.54
2013-01-02	201303	Close	755.25	-26470.59

The mechanism which causes prices to move is the structure of the orderbook. If the orderbook is heavily weighted on the buy side prices will tend to increase. Similarly if the orderbook is heavily loaded on the sell side prices will tend to fall. It then stands to reason that there is economic merit in being able to forecast where the fat of the orderbook will be on any given day.

In order to look for repeatable signals we first have to perform some preprocessing on the data. For each commodity we find the bought and sold event at every date together with the VolumeEst associated with it. We look back five days prior to the event and 10 days after the event. Stacking up all the events enables us to calculate statistics on the days after the event has occured.

2 Event Studies

2.1 Aggregated Example

The facet plot below shows the Bought and Sold event statistics. In this example the data is aggregated over the entire dataset. The x-axis is labeled as index where the index is zero at the day of the event, indicated with vertical dashed lines. Indices \(-2\) and \(4\) represent 2 trading days before and 4 trading days after the event occurred respectively. We stack the percentage moves of each index together and determine the mean and standard deviation of the percentage moves within each index. In the plot below the mean is represented by the solid black line and the standard deviation away from the mean is shown as error bars.

The telltale signs of a trend system are clearly present. Notice that for the Bought (Sold) event there is an increase (decrease) in prices leading up to the event. Recall the trend systems are always late to the party and inevitable overstay their welcome.WeI expected to see the prices increasing (decreassing) after the Bought (Sold) event before returning to some equilibrium value. This would correspond to the trend systems overstaying their welcome. We do not observe this in the graphs, but do see faint signals of this in the table below. We see that when the buy signal is hit mean percentage price changes, up to 10 trading days after the event, are pretty flat. The same seems to hold trye for the sell signals.

In the following we show the disaggregated results for each of the commodities in the argicultural and softs dataset from Bridgeton. In some of the commodities the overshoot effect of a trend system is more pronounced than others, but in general there is not much of a signal.