For example portfolios/sims are buying stocks when there is no 5/21 sma spepscy crossover and not selling stocks when there is a 5/21 sma spepscy crossover.
Your SP500 Market Timing Model bought stocks as did this test port https://www.portfolio123.com/port_summary.jsp?portid=1073445 on 6/29/15 when there was no 5/21 sma crossover. This test sim bought stocks on 7/6/15 when there was no 5/21 crossover: https://www.portfolio123.com/port_summary.jsp?portid=1078112. All 3 of these ports/sims still are invested in stocks even though the 5 sma of earnings has been below the 21 sma of the earnings for several months and all 3 have a weekly rebalance.
I am under the understanding that the false buy signal on 6/29 and 7/6 was due to earning revisions?
I have not noted there to be erroneous signals in the past so did something change in how you derive these estimates?
Why did the ports and sim buy stocks a week apart with the same rule(shouldn’t it be consistent)?
Why didn’t the sims/ports immediately sell stocks after they erroneously bought them as the buy signal has been false (5 is below the 21) for several months and they all have a weekly re balance?
Why is there such a discrepancy between the custom series tool and the point in time tool for spepscy (the 5 wk sma is still below the 21 wk sma on the custom series tool and the 5 wk sma is above the 21 wk sma on the point in time tool)?
Note that if I run a sim based on the above 5/21 rule it has it in stocks but if I start a port with today’s date it has it out of stocks so there is a significant discrepancy not just between these values on the point in time tool and custom series tool but also in how the sims and ports interpret this rule. What is the time frame to fix this?
The custom series to re-create the SPEPSCY series did not match the way we actually calculate it. Had to do with the adjustment for NA’s.
HOWEVER… while inspecting this we discovered a problem with the way we currently do it and we will need to rebuild the SP500 estimate series (in short the adjustment was being applied to the ratio but it should have been applied only to the numerator).
We will rebuild the series shortly and make an announcement. I don’t think there will be a major impact since the values are not much different, but past simulations will change somewhat.
Very sorry about this and thanks for bringing this up.
There still seems to be an issue on the timing of when the sims and screens use the #SPEPSCY data versus the Ports. On 9/19/15 after Saturday’s update, the SMA(5,0,#SPEPSCY) was 117.46 and the SMA(21,0,#SPEPSCY) was 117.49. The SMA 5 was below the SMA 21. This was recognized by the ports on the 9/19/15 rebalance, but not by the sims or the screener. The sims do not recognize this until the 9/26/15 rebalance. The screener will recognize this if screened on Monday 9/21/15.
I get the same calculations today (SMA 5 < SMA 21) on 9/19/15 when looking at the Fed Model, so the data is the same.
Since the data was available on 9/19/15 after the Saturday update, why do the sims and screener not see this data until the following week? The ports have it on Saturday 9/19/15 and recognize the SMA 5 is below the SMA 21. The sims do not. As a result the sims are taking timing trades on the wrong dates. This lag seems to be constant when looking back at the sim, and is a separate issue from the rebuilding of the #SPEPSCY data.
Mike,
You said earlier; “Marco, There still seems to be an issue on the timing of when the sims and screens use the #SPEPSCY data versus the Ports. On 9/19/15 after Saturday’s update, the SMA(5,0,#SPEPSCY) was 117.46 and the SMA(21,0,#SPEPSCY) was 117.49. The SMA 5 was below the SMA 21. This was recognized by the ports on the 9/19/15 rebalance, but not by the sims or the screener. The sims do not recognize this until the 9/26/15 rebalance. The screener will recognize this if screened on Monday 9/21/15. I get the same calculations today (SMA 5 < SMA 21) on 9/19/15 when looking at the Fed Model, so the data is the same. Since the data was available on 9/19/15 after the Saturday update, why do the sims and screener not see this data until the following week? The ports have it on Saturday 9/19/15 and recognize the SMA 5 is below the SMA 21. The sims do not. As a result the sims are taking timing trades on the wrong dates. This lag seems to be constant when looking back at the sim, and is a separate issue from the rebuilding of the #SPEPSCY data. Your help is greatly appreciated. Thanks, Mike”
I just saw the same issue where my live Port’s Hedge sma crossover (9/21) was one week ahead of the sim’s (9/28). I was happy because it enabled me to save money but it does call into question why my port is one week earlier than my sim’s.
Does P123 have any idea why? I can share the port and sim if needed.
I believe it’s because portfolio & screen backtests do not process the last saturday because there are no future prices , so it’s pointless: there are no transactions to recommend since there are no Monday prices, and it won’t use Friday’s prices.
Live systems are different, they give future recommendations using Fridays prices as a best guess for Monday’s prices.
I think people are using backtests to get the latest signals. They are not designed for that. The latest signals should come from live portfolios or running a screen in the present.
I have to check if running a screen in the present. I believe it uses “as-of” Friday price, cutting off the Saturday data point for SPEPS. This should not be so.
Thanks for the reply and for the fantastic work!
If I want to replicate past performance of a system that uses SPEPS into the future in the most exact way is probably better to sell or to buy one week later after the live portfolio recommendation. I do not think that this is crucial but probably achieves a more similar outcome with sims.
I thought I’d wait till today, Monday, to reproduce and fix this, but I’m not able to reproduce.
No matter what I do (sim, screen, screen backtest), the value Close(0,#SPEPSCY) is returning 116.4928, which is this Saturday value for the SP500 Curr Y estimates. Moving averages also seem to include this value.
For a screen, if you add a rule to any screen for close(0,#SPEPSCY)<116.50 in the case of 10/10/15’s value of 116.4928 and run the screen on Oct 10th after the update, or Oct 11th, the screen will return no securities as it can’t see Saturday’s #SPEPSCY data until a screen date of Monday Oct 12th. You can try this now. This is a problem if your screen uses the #SPEPSCY parameter.
For a Sim, if you add a sell rule of close(0,#SPEPSCY)<116.50 in the case of 10/10/15’s value of 116.4928 and run the Sim through today October 16th, it will not produce sell orders for Monday 10/12/15. It should produce these orders as the #SPEPSCY data on Saturday after the update was less than 116.50.
As I mentioned in my prior post, the sims (balanced weekly) do not see Saturdays #SPEPSCY until the following Saturday, causing signals to be a week late, regardless of when the Sim is run. In the case of the screen, it will recognize a rule with Saturdays #SPEPSCY on a Monday Screen date. This is also too late.
Finally found the problem. Live systems were picking up the latest data , but anything historical set on the weekend was picking up the previous weekend.
