Monthly Standard Deviation

[robika]

I'm applying what I would consider a rather conservative strategy: 17620 has a mean of 29.72 with a SD of 2.59. However, this SD is based on yearly compounded returns. If you look at the monthly means and calculate a standard deviation for the monthly returns, the numbers are all over the place. For instance, for the month of February, the mean is 2.5 with a SD of 4.64, which to me means that for this strategy the month of February is very uncertain on returns, ranging from -10.03% to 9.11%. For the strategy that I follow there is only one month where the monthly data could technically be treated as a normal distribution and that's December.

So that being said, I was wondering if there are any strategies where the monthly SDs are a bit more conservative as compared to the monthly mean? I think that I would forego huge returns in hopes of having consistency throughout each month. Any ideas?

[mindofmush]

I have yet to see a daily strategy (CAGR>12%) that doesn't have a negative month.

In order to achieve a standard deviation less than the monthly mean, all monthly returns need to be positive. Maybe you can create a strategy by replacing the uncooperative months (like February) with the always positive G fund.

It looks like a huge time consuming project either way (looking for an existing strategy or creating a new one that fits your criteria).

Good luck.

[TopComm]

I have yet to see a daily strategy (CAGR>12%) that doesn't have a negative month.

In order to achieve a standard deviation less than the monthly mean, all monthly returns need to be positive. Maybe you can create a strategy by replacing the uncooperative months (like February) with the always positive G fund.

It looks like a huge time consuming project either way (looking for an existing strategy or creating a new one that fits your criteria).

Good luck.

Concur. Strategy 68 uses the G fund for the entire year. From 2004 - 2016, the average monthly return is .253%. This has a standard deviation of .101. Based on a monthly return, I doubt there will be anything better as any of the other funds can have a really bad month or a really good month which would throw off the standard deviation.

[Bosch]

The tspcalc strategies that I’ve seen discussed the most here all have at least one negative month every year from 2004-2016. Examples are 16517, 16384, 16198, 17680 and 18390 have all had at least one negative month from 2004-2016. And they all had a negative month prior to December of each year. But 2017 has been different. Each of these strategies has been positive from January thru November. So far December is the only negative month. One other is 16999 which only had one year without a negative month from 2004-2017, that was 2006. Whether December ends up positive or negative, it will be an anomaly for these strategies at least.

[evilanne]

Let us know what you come up with robika. You can play with the IFTs and try to reduce variability for each month but with the limitation on IFTs, your expected return is likely to decrease proportionally to the amount of risk that you eliminate.

Moving to G on the last day of January with no IFTs in February only reduces the mean by about 2% and more than doubles the SD http://tspcalc.com/seasonal.php?ID=19699&years=04-17 The one month that had a significant loss of >10% in February was in 2009 when the market was was bottoming out during the last recession, but your strategy had over a 30% return that year. A significant drop could happen any time when you are in the market, but what are the odds that it will happen ever again in the same way? It all depends on your risk tolerance. Are you willing to change your strategy based on a single bad monthly datapoint?

[robika]

Let us know what you come up with robika. You can play with the IFTs and try to reduce variability for each month but with the limitation on IFTs, your expected return is likely to decrease proportionally to the amount of risk that you eliminate.

Moving to G on the last day of January with no IFTs in February only reduces the mean by about 2% and more than doubles the SD http://tspcalc.com/seasonal.php?ID=19699&years=04-17 The one month that had a significant loss of >10% in February was in 2009 when the market was was bottoming out during the last recession, but your strategy had over a 30% return that year. A significant drop could happen any time when you are in the market, but what are the odds that it will happen ever again in the same way? It all depends on your risk tolerance. Are you willing to change your strategy based on a single bad monthly datapoint?

I was playing around with the calculator a little yesterday and realized that it will definitely be a challenge, but hey I got the rest of my working life to tinker with it, right?
Anyway, I think I'm going to probably try to focus on more recent years, or maybe exclude 2008 & 2009 since they're such oddball years. I know that those years are part of what can happen, but they are more the exception than the rule. I'll post on here ones that I "find."

[12squared]

I'm applying what I would consider a rather conservative strategy: 17620 has a mean of 29.72 with a SD of 2.59. However, this SD is based on yearly compounded returns. If you look at the monthly means and calculate a standard deviation for the monthly returns, the numbers are all over the place. For instance, for the month of February, the mean is 2.5 with a SD of 4.64, which to me means that for this strategy the month of February is very uncertain on returns, ranging from -10.03% to 9.11%. For the strategy that I follow there is only one month where the monthly data could technically be treated as a normal distribution and that's December.

So that being said, I was wondering if there are any strategies where the monthly SDs are a bit more conservative as compared to the monthly mean? I think that I would forego huge returns in hopes of having consistency throughout each month. Any ideas?

Daily 19403 is my best attempt at maximizing Mean/StdDev ratio on a monthly basis. It was compiled from several popular strategies after pasting the month by month data for 2004-16 into Excel. As of 11/29 it had the highest Mean/StdDev for each month, excepting May (0.92 vs 1.22 for 18390) and August (0.80 vs. 0.90 for 16922). However, its annual Mean/StdDev is only 3.39. I suspect that this is because positive months aren't balanced out by negative ones as often as in the high sigma strategies.

[robika]

I'm applying what I would consider a rather conservative strategy: 17620 has a mean of 29.72 with a SD of 2.59. However, this SD is based on yearly compounded returns. If you look at the monthly means and calculate a standard deviation for the monthly returns, the numbers are all over the place. For instance, for the month of February, the mean is 2.5 with a SD of 4.64, which to me means that for this strategy the month of February is very uncertain on returns, ranging from -10.03% to 9.11%. For the strategy that I follow there is only one month where the monthly data could technically be treated as a normal distribution and that's December.

So that being said, I was wondering if there are any strategies where the monthly SDs are a bit more conservative as compared to the monthly mean? I think that I would forego huge returns in hopes of having consistency throughout each month. Any ideas?

Daily 19403 is my best attempt at maximizing Mean/StdDev ratio on a monthly basis. It was compiled from several popular strategies after pasting the month by month data for 2004-16 into Excel. As of 11/29 it had the highest Mean/StdDev for each month, excepting May (0.92 vs 1.22 for 18390) and August (0.80 vs. 0.90 for 16922). However, its annual Mean/StdDev is only 3.39. I suspect that this is because positive months aren't balanced out by negative ones as often as in the high sigma strategies.

Interesting. I came up with 19717 yesterday using the same technique. Couple of minor differences between the two: I think 19717's February numbers are little better (SD and subsequently it's range is smaller), and November/December are pretty much on par with yours (yours goes to F, mine to G). I'm going to combine the two and see what it comes up with.

[12squared]

I came up with 19717 yesterday using the same technique. Couple of minor differences between the two: I think 19717's February numbers are little better (SD and subsequently it's range is smaller), and November/December are pretty much on par with yours (yours goes to F, mine to G). I'm going to combine the two and see what it comes up with.

19717 & 19771 make great improvements to February - taking different approaches from most others. I find it remarkable that their lowest annual returns from 2004-2016 are only 1.4 & 1.2 SD below the mean, respectively.

Here are the Daily strategies I've found which have the highest monthly Mean/StdDev

January: 19403, 19771
February: 19771
March: 19403, 19717, 19771
April: 19403
May: 16319, 16262
June: 19403, 19717, 19771
July: 19403, 19717, 19771
August: 16922
September: 16922, 19403, 19717, 19771
October: 16922, 19403, 19717, 19771
November: 19403, 19771
December: 19157 (note IFT to G on 20th trading day, not last)

[skiehawk11]

I wonder if tspcalc could do the following calculation.

I'd like to take a strategy with the highest return from 2004 to 2005 and see how it's performed from 2006 - 2017. Is that possible?

[Travis1984]

I wonder if tspcalc could do the following calculation.

I'd like to take a strategy with the highest return from 2004 to 2005 and see how it's performed from 2006 - 2017. Is that possible?

http://tspcalc.com/seasonal.php?ID=16753&years=04-05

The above link is top strat from 2004-2005

http://tspcalc.com/seasonal.php?ID=16753&years=06-17

This is it from 2006-present.

Amazing really...

[Travis1984]

Here I’ll do it better. Here’s the top strat from 2004.

Extrapolated then to the full run of data...insanity.

http://tspcalc.com/seasonal.php?ID=1971 ... sortby=avg

[12squared]

I wonder if tspcalc could do the following calculation.

I'd like to take a strategy with the highest return from 2004 to 2005 and see how it's performed from 2006 - 2017. Is that possible?

I've found that using 2004-07 gives more consistent results going forward.