We have now made available to our subscribers dozens of economic activity series from the Federal Reserve’s Economic Data (FRED) for use across our website. The complete list of series is in the Factor and Function Reference under MISC / TIME SERIES IDS. Much of this data goes back much further than twenty years.
In addition, these series being introduced in this release have a PIT version (though the pre-existing FRED series like ##UST10YR don’t). So if you want to use the Real Gross Domestic Product in a screen or simulation, you can use either the data from the revised series (e.g., [font=courier new]Close(0, ##RGDP)[/font]) or the point-in-time series (e.g. [font=courier new]Close(0, PITSeries(##RGDP))[/font]).
A potential pitfall of using the series as point-in-time is that there can be significant jumps when series are revised. To see a demonstration of this behavior, try using Fundamental Charts to plot the past 10 years of [font=courier new]Close(0, ##RGDP)[/font] and [font=courier new]Close(0, PITSeries(##RGDP))[/font].
If you need to download any of these series, you can do so at https://fred.stlouisfed.org/ or by using an Aggregate Series with a rule like [font=courier new]UnivAvg(“True”, “Close(0, ##RGDP)”)[/font].
We are working on phasing out the Compustat time series for economic indicators. All the non-economic indicators in that series already have FRED equivalents in our system (e.g. ##UST10YR and ##FEDFUNDS). Now the economic indicators have FRED equivalents too. At some point in the near future we will be replacing your existing commands using Compustat time series with the FRED equivalents (along with the appropriate multiplier/divisor if series are equivalent except for a 100:1, 1000:1, or 1:1000 ratio).
On the one hand, using these series is easy. To get the most recent reading, just use Close:
That should work in any technical rule that has the series parameter. For example, if you want to take a 10-period simple moving average, that would be:
The issue, of course, is that it is a 10-period simple moving average. So you need to know what the periods of the time series are. In this case, real gross domestic product is quarterly, so the 10-period average is over the period of 2.5 years. Write your rules accordingly.
When you mean, PIT data, are you refering to the non-revised data, or real PIT data?
I mean, for example, unemployment is published the first friday of the following month. So the data of January is published the first friday of February. When you download FRED data, the date asigned to that realeased is 1st of January, so you need to calculate the real date of the release.
Could you explain further how the PIT works in P123 for the FRED data?
When we say PIT data, we mean the entire revision history is represented, so it can be observed in the state it would have been for any given date. For FRED PIT series, the values are provided with effective dates when the series is retrieved, so our PIT series should exactly represent what is available on the FRED API.
To take advantage of the PIT series, you have to use the PITSeries function as detailed in the original post. Here’s a demonstration for PITSeries that Yuval came up with:
[font=courier new]SMA(4, 0, PITSeries(##RGDP)) * SMA(12, 0, PITSeries(##CPI))[/font]
“The use case for this is to adjust minimum market cap or minimum sales in universes for long backtests by this factor. So if your minimum market cap is $100 million in 1999, you would use this series to calculate the minimum market cap in 2020.”
It generally makes more sense to use revised data for series based on sampling and polls (RRSFS, HOUST, UNRATE). Their average revisions are smaller now than 10 years ago or more (it’s a fact), because tools and methodologies have improved (it’s my interpretation). PIT data consider noise due to old measurement errors is a part of the information. I have looked into this thanks to ALFRED legacy data series for the above-mentioned series. I also noted that UMCSENT (poll) is not revised. In the case of INDPRO (sampling), revisions have sometimes been large enough in the last few years to occasionally change a trend calculation. INDPRO revisions may happen a long time after the 1st release (more than 6 months), so I prefer ignoring this series.
Yuval and P123 team thanks a lot for including more FRED data and giving the choice of PIT or not.
For most series, the revised series will work just fine, but for the following five series, I recommend using the PIT series instead of the revised series to get them to work properly historically.
Revisions on these five series incrementally discard old values such that the revised series may not work as desired if historical values are needed (e.g. SMA):
##INV2SLS (Total Business: Inventories to Sales Ratio) has had values back to 1948-01-01, and the latest revision has values back to 1992-01-01. The maximum supported offset for continuous usage is 111.
##INVTOT (Total Business Inventories) has had values back to 1980-12-01, and the latest revision has values back to 1992-01-01. The maximum supported offset for continuous usage is 111.
##ORDERSCAP (Manufacturers’ New Orders: Nondefense Capital Goods ex.Aircraft) has had values back to 1968-02-01, and the latest revision has values back to 1992-02-01. The maximum supported offset for continuous usage is 109.
##ORDERSDUR (Manufacturers’ New Orders: Durable Goods) has had values back to 1958-02-01, and the latest revision has values back to 1992-02-01. The maximum supported offset for continuous usage is 110 (starting 1999-08-04).
##RPCE (Real Personal Consumption Expenditures) has had values back to 1959-01-01, and the latest revision has values back to 2002-01-01. The maximum supported offset for continuous usage is 7. (There is a large revision around 2004 that causes this.)
For these five series, it is also recommended that you plot your technical functions on Fundamental Chart to ensure that they work properly.