Forecasting

This guide walks you through forecasting a pairs strategy, and understanding the output of the forecast model to make better decisions about which pairs to trade.

This tutorial is split into 3 sections:

• Comparing forecasts across multiple pairs
• Viewing forecast distribution of a single pair
• Understanding edge

The forecast model will generate forecasts along different points of the distribution curve. There are primarily two ways to view this data without information overload - by comparing a single forecast path across multiple pairs or looking at the entire forecasted distribution for a single pair.

Comparing a single forecast path across multiple pairs

A forecast path is a particular percentile on the forecast distribution. You can turn on or off different paths by checking the percentiles at the top of your 'Forecast' tab. A natural starting point is to compare the median, or 50th percentile, of multiple stocks. Make sure only the 50th percentile is checked and highlight multiple pairs by holding CTRL. The example below illustrates comparing the medians across two pairs: AMD / INTC and MSFT / GOOG. These are both technology pairs, but the former is a hardware-related sub-industry and the latter is a software-related sub-industry.

(click for larger view)

Viewing full forecast distribution for a single pair

In this example we're looking at all the percentiles for a particular pair. You don't want to select multiple pairs here because it'll plot too many series & create information overload. To do this, make sure all the percentiles you want to look at are checked on at the top of the page. Then select a particular pair. The screenshot below shows the forecasted return distribution for AAPL / MSFT.

(click for larger view)

Understanding edge

Viewing the forecast distribution naturally begs the question, "How much am I likely to make on this pair?"

Other models in the PairsTrader System is trying to assess whether a particular trade is a good opportunity or not. The forecast model, on the other hand, is about answering questions like, "If I trade this pair, how much can I make, probabilistically?" However good or bad a certain trade is, no model or trading system is omniscent and any individual trade has some probability of making money and some probability of losing model.

Pairs trading is all about consistency - and that means taking positions where you're more likely to win than lose. To do this, you need to take positions where the amount received when you win is more than the amount you give up when you lose. That's house Vegas does it and it's called edge, advantage, or "house odds".

You can measure how much edge you have in a trade by looking at edge statistics in the PairsTrader System.

There are 3 edge statsistics:

• Low probability edge (LPE)
• High probability edge (HPE)
• Extreme probability edge (XPE)

The 'probability' in these measures of edge refers to how much probability mass is required to warrant the calculated edge. Low probability edge is calculated as the 75th percentile forecasted return less the 25th percentile forecasted return. So the probability mass required to push away on both sides of the median is 25%. These events have a high probability of occuring.

High probability edge is the difference of the 90th percentile less the 10th percentile. You're requiring outcomes to have a higher probability, 40%, of being away from the median. These events are less likely to be observed.

The XPE is the difference between the returns for the 95th and 5th percentiles. 45% probability mass is required for outcomes from being away from the median. These events are even less likely to happen.

You should always think of edges as spreads, with LPE as the spread measuring events closest to the forecasted median. if the model is right, these events have the highest probability of being observed and high LPE values indicate a very good pairs trading opportunity.

If you swing for the fences you might want to pay attention to HPE or XPE - these values will typically be higher but have less chance of being realized. When realized though, you'll win in big ways.