Jessica, this is interesting. I have a lot of investment experience (and that is investing as opposed to Quantopian-type trading) but little programming experience. I remember the Billion prices project but haven't kept up with it. I thought State Street acquired it. Anyway, are you taking care to lag the CPI to account for the fact that its release comes well after the period for which inflation is being measured? From a fundamental perspective I would expect the source of inflation to impact which sectors benefit. An oil price shock will be different from a rise in labor costs. Also, markets probably react differently to expected from unexpected inflation. Theoretically, markets have discounted expected inflation. It used to be difficult to measure expected inflation. Now with TIPs, a simple estimate of inflation expectations is the yield of a nominal bond less the TIPs yield. I think it would be interesting to see how the sectors react when the real time inflation rate is exceeding the expected rate and vice versa. The implementation would still be limited to inflation in the short run. The real time series gives you a sense of recent prices changes. Thus you would might compare this recent inflation information with the expectation for inflation over the short term (using shorter maturity nominal and TIPs bonds). However the markets may care more about longer term inflation which is not going to be captured by the real time series.

If anyone is still reading, I'd add that the idea of using real-time date (in this case prices to estimate inflation) is fascinating. One might be able to predict surprises in government (and industry or even company) releases. From a trading point of view one would want to be able to predict the value that is going to be released with better accuracy than the consensus. One would trade when one's model produces an estimate different from the consensus estimate which would constitute a surprise and should move markets. I'm thinking of data series like housing starts, permits, prices; auto sales etc. Could one get an idea of this from web traffic; searches; changes in number of listings etc. It sure would be nice to have access to big data from firms like Google or Amazon. I'm pretty sure some hedge funds are on top of this. There's a thread on trading based on Google Trends data which is along these lines.

Hi Rex,

Thank you for your reply - most importantly, thanks for catching that I had actually NOT lagged the data series in my test so far.

I went back and lagged the BLS monthly CPI by 20 calendar days and then ran the same backtest using that as the signal - see backtest attached . This actually looks a lot more like what I expected, which is that it would be more profitable to use the real-time signal than to wait on the monthly number.

In terms of using the real-time indicator, the BPP has spun out a company called Price Stats who maintain the index and license/distribute through State Street via a couple of options - the most expensive being to get the true real-time data. I haven't verified this, but it looks to me like they do make the index publicly available at a 10 day lag, which is shorter than the lag for the monthly number. I will post another backtest showing the results using the daily index + 10 day lag as well.

Your train of thought on predicting surprises in the government releases is exactly the direction I was thinking would be interesting to take this work. Though the literature on State Street's site (and what I heard from Prof Rigobon) both indicate that their intention with this data is NOT to provide a forecast for the next release - you have to imagine that is a big part of why someone would pay for it.

Last note - I know there is competition in this space, there's at least one other company collecting primary data on global inflation - but I haven't had a chance to try to get sample data to do a comparison (see: http://www.premise.com/). And of course as you point out you could take this as general as the Google Trends idea.

Thanks again for your comments and insight, Jess

Here's the same strategy using the daily index (as in the first post) but with a 10 day lag, so simulating making the data available to the algo when the free version of this data would be expected to be publicly available.