Choosing a Seasonal Investing Strategy

[ewok55]

Ewok 55, I'm pretty sure I did what your are describing. Although I started out with a different goal I think I ended up testing what you wanted to see.

Below are the seasonal mixes and yearly returns every year from 1989 to 2016 using what I called the "Max" approach. I called it "Max" because it uses the maximum number of years (available at the time) to come up with a seasonal approach.

Max Mixes Capture.JPG

1989 is based on 1988.
1990 is based on 1988-1989.
1991 is based on 1988-1990.
...and so on until...
2016 is based on 1988-2015.

This is what the "Max" column in my earlier posts were based on.

I agree that this shows more realistic returns because it is based only on information that would have been known at that time (but as a posted above I don't think it's wrong to look at hypothetical past returns if you understand their limitations). However, you can see that the seasonal mixes started settling down and changing less as time went on. This seems to validate what the seasonal approach is based on--that the market has patterns that tend to repeat.

This kind of goes back to my original post where I asked if someone was using something other than past monthly averages to come up with a seasonal strategy. Because all the seasonal mixes mentioned in these forums all look so similar it seems like they are mostly using past monthly averages. I asked this question because if someone is using something else to create their mix then maybe back in 1989 they would have been in a better position than merely relying on what happened in 1988.

If anyone is interested, incorporating what happened in 2016, if you were following this approach this year January would have changed from S to F and March would have changed from C to S (although both of these months are very close between the two funds that switched).

If anyone sees mistakes let me know. There was a fair amount of copying and pasting formulas going on.

This is the same thing I'm looking at... I got up to 1998 earlier today and plan on continuing. It has been a huge help for me to understand the data better and get a sense of normal market fluctuations (that otherwise might have scared me!). One thing I've noticed is that there are definite trends, but they do change over time (case in point, the G and I funds used to be much better for some months). I plan on looking further into this, possibly coming up with something that utilizes other market trends as "triggers" for different seasonal strategies ... very fun if you like statistics . Thanks for sharing the data, Aitrus!

[gclapper]

Hello all and please feel free to correct me. I have been trying to follow this thread and it seems to me that everyone is talking about the same exact thing but going around and around confusing the issue. Here is how I understand it, how I have used Aitrus’s spreadsheet (with tweaks of my own), and used Aitrus’s Seasonal Thread teachings. FYI, I just started looking at this since November 2016 so I have not even had a full year yet to see how it goes; but very hopeful and way more than I was doing.

1. Determining a seasonal strategy takes the max number of years that there is data available for (at that time) and using that data to determine the following year’s strategy. So use 1988 through 2016 data to come up with seasonal strategy for 2017. This would stand true if it was 2001 in which this method would use data from years 1988 through 2000 to come up with a strategy.

2. At the end of each year you add that year’s data to the overall data pool and reevaluate your seasonal strategy. This may cause little tweaks to adjust to changing CAGRs, PNR, etc. So strategies would change and evolve over the years. Each year’s impact will gradually decrease as more and more years are added to the data pool.

3. When we are looking at a strategy such as my “gclapper M3’ it will be reevaluated each year. This year I am sticking with it but I did notice that there could be a slight change to it with the 2016 data but decided it was not significant enough to warrant a change for a slightly increased return potential that brought with it a slightly increased risk.

4. When we are looking at Aitrus’s spreadsheet or reporting that a seasonal strategy resulted in returns for each year 1988 through 2016, it does not mean we would have used that strategy those years (refer to item 1 above; data pools would have been increasingly smaller going back each year). To me showing how a strategy would have performed on each year back to 1988 is just another visual representation of how the overall data from 1988 through 2016 performs on average. It is just a visual/alternate way to show the fund trends over the years and into each month. I think everyone was saying this but in different ways as to confuse the other.

5. The reason why a lot of the strategies are starting to look alike is because the number of years that we have data for is starting to smooth out the trends; as Aitrus stated previously. I think that back in the early years 1988 to 1990s would have been a lot rougher to get a seasonal strategy that was consistent or statistically significant. Other investing strategies and knowledge would in my opinion been necessary to get good returns.

6. Monthly data is using the Cumulative Annual Growth Rate (CAGR – aka: average rate of return) to determine how each fund performed each month and over the years. CAGR is different from a straight mathematical average and takes into effect negative compounding. Positive to Negative month ratios are also used to help determine the risk associated for each fund during each month over the years that we have data for. PNR combined with CAGR and each person’s risk level preferences determines what each person’s seasonal strategy will be.

7. Minimizing loss is very important as Aitrus stated in his Seasonal Strategy Thread. Over the long term these minimizations really have a huge effect on your ending balance when you retire. Did some Excel magic to prove this to my self and it is really scary that even small losses can take years to gain back sometimes.

So basically in my humble newbie opinion the seasonal strategy methodology is very simple. That is the reason I like it and makes sense to me. The discussion in this thread is all great info but I think that we are all saying the exact same thing; just in different ways that end up with the exact same results. If I missed the mark then my bad and ignore my post…lol.

[Aitrus]

You're welcome guys, and thanks for the validation as far as the methodology goes. I was kind of loopy last week - working a lot of extra hours and on meds to combat a head cold, so I'm not entirely sure what I was thinking in a couple of my posts. I'm just glad that evilanne always seems to always be there to have my back.

[Aitrus]

Good points, gclapper.

It's interesting that we all seem to be stopping at 1988 for backtesting. I think that's mostly because that's how far back TSP has data for the C, G and F Funds. If it were 1990 and I was coming up with this strategy at that time (at the ripe old age of 10), then I would have pulled data going back for as far as I could find it and could validate it's reliability. Say 1960 - 1988, for example. I would still be in the same position we are in today: using past results to come up with a method for tomorrow.

The only reason I haven't gone back to years before 1988 yet is because I haven't done the digging to find the data to represent the TSP Funds if they had existed prior to 1988. Simple as that. If anybody has it on hand, I'll gladly incorporate it into my spreadsheet. More years = better reliability of the system.

[evilanne]

I'm not sure going back further would be fruitful. So much has changed over time. The S & I funds were added to the TSP in 2001, L Funds added in 2006 which have different weights in the primary funds over time. I think it was the mid-late 1990's when you could view your account online, they allowed online IFTs followed by the limitation on IFTs that led folks to looking at seasonal methods. The technology we now use for managing our TSP accounts has changed significantly.

Our economy has fundamentally changed, so looking back at older data may actually be counterproductive. If you want to do more detailed analysis, you probably need to look at more macro economic & political trends. I know there is a current trend for everyone to use "big data" to analyze everything but I'm not sure that more data is always useful, that analyzing more data is worth the effort, or that we have access to enough detailed data to make it any more useful than what you have with the seasonal data here. It never hurts to try and build a better mouse trap...just don't let all the mice escape in the process & keep investing

[KWG_5]

I'm reviving my old topic because I have an updated tables that include 2017 and I looked a little bit at daily seasonal strategies. The purpose of the table was to see what results you could get based only on what you knew at the time. Simultaneously it looks at whether considering more or fewer past years was better. I just have concerns about the viability of investment strategies that are created after-the-fact using information that wasn't available at the time. I think this table shows 100% realistic strategies and their performance.

It's a little confusing to explain, but I'll try summarize here so you don't have to go back to the start of the thread. What I did was run 30 different scenarios looking at what returns someone would have made if they based the investments on a monthly seasonal strategy looking back at the past 1-29 years. I only had monthly data going back to 1988 so the longer the time span considered the fewer years I could calculate. Seasonal strategies were based solely on which fund performed best (on average) for each month for the given number of years. This means that strategies can change each year based on the new information obtained from the previous year.

For example, the 10-Yr column means that each year the previous 10 years were looked at and the fund that performed the best each month (on average) was used to invest in for that month for the coming year. Each new year the same process was redone.

The second column to the right, "Max", keeps using more years to choose a seasonal mix. The first year it only used one year. The second year it used 2 years and so on until the 29th year it uses 29 years.

For "Moving Averages.jpg" the percentages are the returns. For "Moving Averages Minus C Fund.jpg" the percentages are the difference between the yearly gain of the strategy and the C Fund. A positive value means it outperformed the C Fund (or S&P 500). A negative value means it did worse than the C Fund.

I'm not sure how helpful these charts are, but to me it shows that basing investment decisions on only the past few years has not been too effective in recent history (using the strategy in the 5-Yr column would have been an awful strategy). The "Max" column seemed to be the best performer and it was based on using every year available at the time to choose which fund to invest in. The "Max" strategy didn't have amazing returns but did generally outperform the C Fund.

I also noted that the 13-year averages stood out as consistently outperforming the C fund. I don't know what would be special about 13 years. It's probably an anomaly, but I thought it was interesting.

The farthest right column is new this year. In 2005 I used the top seasonal strategy in 2004. In 2006 I used the top seasonal strategy for years 2004-2005 and so on until 2017 I used the top seasonal strategy that included 2004-2016. This strategy outperformed the C fund 11 out of 13 years and by greater margins than the monthly strategies.

Like I said, it can be confusing. Let me know if you don't understand.

[mjedlin66]

You look like you are having fun. I remember doing all of that in excel when I built my first daily strategy.

I appreciate your analysis. Especially the effort to determine what our window should be. 13 years surprises me, I would expect it to be less than 10 but that is just a hunch that the market of today is significantly different than 10 years ago.