1) How do I interpret the graphs?
The main line in each graph can be thought of as the probability assigned by the model that the current month is in within a recessionary spell. The grey areas represent historically recorded recessions.
2) What's the difference between the two graphs on the "Current Nowcast" page?
Only the time scale. The top graph shows a longer window of time, while the bottom one shows the same data, in a narrower time frame.
3) What model is used to calculate the probabilities plotted?
The nowcasts are produced by a Learning Vector Quantization model similar to Giusto and Piger (2017).
4) The long run graph indicates an outstanding accuracy and timeliness (at least for the time window shown). Is this for real?
Yes it is for real, but no, it is not in real-time. The model - like a few others - performs exceptionally well in identifying recessions, when given the final release of the data. That is the case for all recessions shown in the top panel. The challenge is to obtain a good performance on early releases of the data: this means that you should not extrapolate the past performance displayed in each graph into the present.
5) What is the meaning of the colored bar at the end of each graph?
Taken at face value, the probabilities from the model would result in many false signals. For this reason, two filters are applied to the raw probabilities: first, the probability must be above/below a threshold represented with a horizontal red line in the graphs (typically set at the value 80%). The second filter consists of requiring that the probability is above said threshold for one (yellow bar signal), two (orange bar signal), or three (red bar signal) consecutive months.
6) The colored bar skipped a step (e.g. last month it was green and now it is orange/red). Is that a mistake?
Not necessarily! Most likely this is a consequence of data revisions: it is entirely possible that the model outputs a low probability for a given month, based on the available data, and a higher probability for the same month at a subsequent date, once data revisions are published. That said, while I will do my best to make sure that the results of the model are faithfully published on this site, I take no responsibility for any mistake in these calculations.
7) How reliable are these signals?
This model produces fairly reliable signals. In a real-time test of the past 50+ years, this model produces the following results. When calling peaks, the yellow signal has a 3.83 false positives rate: it signals incorrectly the beginning of a recession almost 4 times as often as a correct signal; the orange signal has a false positive rate of 2, and the red signal has a rate of 0.29. When calling troughs, a single month below the threshold has been correct every single time in the record, two months below threshold have a false negative rate of .67 (i.e. it signals incorrectly the end of a recession 2 times out of 5), and three months below threshold have a false negative rate of 0.2.
8) Which signal should I rely on, then?
That depends on your preferences! Clearly, there is a trade-off between safer signals versus timeliness: which signal you should rely on depends on your views on this trade-off (how much precision are you willing to give up for extra timeliness?).
9) How timely are the signal?
It obviously depends on the signal you choose to follow. The yellow signal is very timely and very unreliable, the orange signal is quite unreliable and somewhat timely; and the red signal is the most reliable but it necessarily implies a lag of at least three months (or more!) before calling a peak.
10) When are the probabilities updated?
I try to update the probabilities at least once a month.
11) Which time series are you referring to?
Total Non-Farm Payrolls (ALFRED code: PAYEMS), Industrial Production Index (INDPRO), Real Manufacturing and Trade Industries Sales (CMRMTSPL), and Real Personal Income Excluding Current Transfer Receipts (W875RX1).
12) Isn't forecasting the present an oxymoron?
No. These are actually forecasts of what the Business Cycle Dating Committee will decide about the current month in a few cycles of data releases. Since the object of the forecast relates to the present, this exercise is known as nowcasting, but it is still a forecasting problem.
13) Can you do this for Canada?
Not really. Unfortunately the collection of real-time data in Canada is still too limited in time to meaningfully test these sort of models for Canada. Yet.