LRC Margin Propagation Thoughts

Repost from KSHB blog:

I understand that the patterns have some variability. I have seen it. I think my concern is two fold.

We hear this cycle "averages" about 50 days. It can be 47 or up to 56 days. Let me propose a scenario.

Cycle 1 – First event on the first day or DAY 1 cycle 1

Cycle 2 – Repeat of first event one day later than 50 days. 51 days.

Cycle 3 – Repeat of second event four days later than 50 days. 55 days.

Cycle 4 – Repeat of third event two days sooner than 50 days. 52 days

Cycle 5 – Repeat of forth event 7 days later than 50 days. 59 days.

So…what was the event starting on the first day of the first cycle now would show up as the same event on day nine of the fifth cycle. Nine days out of sync.

Look at this very real scenario – Help me understand two questions….

Based on the variability, what could be predicted for Cycle 6? Based on the calculations and extension of the variability, it would be about nine days off the “average” cycle, but within the range presented of 47-56 days. Unless you are keeping a total of plus or minus days as a running count cycle to cycle, you would never be able to related the event of the fifth cycle to that of the first. Without explaining this method, people would think you would be nuts to assert it is the same event with an average cycle duration. Maps would not verify.

So, this answers the question that Notes presents: “Why not run out the forecast 6 months if the cycle is known?” For this very reason. The extension of a variable into subsequent cycles makes longer range forecasting problematic based on this method as the variability range would continue to broaden further and further beyond any usefulness.

This is why Gary goes no further out than 45 days. Within that range, you are dealing with a much smaller error propagation since the time duration is within one cycle’s time.

Second – How do you prove out with maps this event with each cycle? You will have problems. Because you have a nine day variability with this cycle event, the margin of error is too great to have statistical significance. Because map analysis is somewhat subjective to start, adding in a range of nine days to find a similar subjective pattern in each cycle position is not good enough to prove this out.

This said, based on this method of analysis, it will be very difficult to prove. However, from a broader perspective in pattern recognition, if you see an event [any event..not just weather] that proves to have specific characteristics that can be identified – happening every 50-59 days – this is enough to provide evidence of a pattern or cycle.

I think the question becomes – how do you prove this, and how do you make it operationally effective in long range forecasting? I think Gary is showing how it can be used in a 45 day window, but because of the variability and seasonal changes – there is a reason why the forecast is somewhat vague and cannot go further out.

I think the critics wanting to push out the forecast even longer or with more detail is somewhat baiting based on the thoughts above. I-for one- have fallen into the trap; hook, line and sinker. I thought initially it should be easy to run out forecasts if the cycle was tighter. I have had marginal success. I don’t think the cycle has as much variability as Gary does, but I still run into the same problems because of the granularity I go to.

All said, I guess it depends on what you want to do with the cycle. If you need trends – this is a great tool. If you are seeking statistical proof of the cycle’s existence…you will need to wait awhile as the current method does not seem robust enough yet to provide that data.

Perhaps further analysis, surface trend blending and other methods could be used to bring numerical proof.

Maybe.