Lootman wrote:IanTHughes wrote:3 years is too short a period for producing any serious conclusions about any Investment Strategy
Since WW2, the average bear market has lasted 14 months. The longest lasted less than 2 years (2000-2002 and 2007-2009). There have been ten bear markets since WW2 so on average we see one every 7-8 years.(*)
The average bull market lasts 4.5 years although the current one has been going on for over a decade now.
So any rolling 3 year period does not guarantee a gain, but you'd be unlucky to invest for 3 years and spend the entire time in a down market. That hasn't happened since 1929. So any 3-year period since WW2 would have either included periods of both rising and dropping prices, or would have seen only rising prices.
I'm sure I've noticed periods of rising prices within a bear market, and periods of falling prices within a bull market - it's not that they don't exist, just that they're dominated by periods of prices moving in the opposite direction. So I think that if you look carefully, you should actually find that a simpler, more categorical statement is true: within any 3-year period at all, there are periods of both rising and dropping prices. But perhaps you meant something like "
So any 3-year period since WW2 would have overlapped with both bull and bear markets, or only with a bull market."? That does indeed follow directly from what you said before it, and both possibilities it mentions do actually happen.
More importantly, as a response to what you quote Ian as saying, both what you said and my modified form of it are rather off the mark. If you're comparing two investment strategies over an N-year period, your most basic result is their relative start-to-end performance over that period. Strategy A's performance is a factor (*) of (end value of A)/(start value of A) - it's gained you money if that factor is greater than 1 and lost you money if it's less than 1. Similarly, strategy B's performance is a factor of (end value of B)/(start value of B). And strategy A's performance relative to strategy B is a factor of (performance factor of A)/(performance factor of B) = ((end value of A) * (start value of B)) / ((start value of A) * (end value of B)) - it's greater than 1 if strategy A has outperformed strategy B and less than 1 if it's underperformed.
That only depends on the values at the start and end of the period you're looking at - so the thing to look at for that primary indication of outperformance vs underperformance is what happened overall from start to end of the N-year period. And if you want to know how reliable an indicator you're getting of how the strategies' performances compare over the long term, you need to look at many different N-year periods and see what proportion of them give you the right answer. I.e. what is primarily important for that purpose is the question of
how unlucky you need to be to get the wrong result from looking at an N-year period, and the way to study that is to look at the range of outcomes you get from rolling N-year periods. The details of how a particular outcome is achieved may satisfy one's curiosity, but whether e.g. strategy A outperformed by 10% over 3 years by outperforming at a fairly steady 3.23% per year during an uninterrupted bull market or by outperforming by 15% per year during 2 years of bull market and underperforming by 16.82% during a year-long bear market in the middle don't make any difference to the question of whether 10% per 3 years is a good estimate of how the strategies compare long-term.
Of course, one of the problems with that is that one doesn't in general know what the right answer is about which strategy is better in the long term - if one did, one wouldn't need to compare the strategies in the first place! So one doesn't know in general what percentage of the rolling periods are giving the right answer. But if the answers given by the rolling 3-year periods are split with say 80% of those periods saying strategy A is better and 20% that strategy B is better, one does know that the chances of a randomly chosen 3-year period giving the right answer are at most 80%. In that case, I don't think I would say that a 3-year period was long enough to produce a "serious conclusion" about how the strategies compare - a "tentative conclusion", yes, but there would be too much of a risk of it being wrong for me to regard it as being wrong.
A final issue about this is that if strategy A is the one you're running and you're comparing it with a benchmark strategy B, you generally can get data about strategy B that covers a large range of rolling 3-year periods, but you don't have equivalent data for strategy A (if e.g. you have 20 years of data on how strategy A has done, why on earth are you trying to conclude anything about comparisons over 3-year periods when you can do a comparison over a 20-year period?). So you have to look at whether a 3-year comparison is good enough for arriving at a serious conclusion about how strategy P compares with strategy B, where P is the best proxy you can get for strategy A for which you can get data about a similarly large range of rolling 3-year periods.
To give a concrete example: suppose that I'm a HYPer who wants to know whether my HYP strategy A would be better replaced with one that avoids high yields and is otherwise similar, and am wondering whether comparison over the 3-year period I've held so far is long enough to justify serious conclusions about which is better. I don't have enough data to cover more than one 3-year period of strategy A, so I choose a proxy strategy P for which I can get enough data to cover a lot of rolling 3-yesr periods. A reasonably obvious candidate for P is investing in an idealised FTSE 350 Higher Yield tracker: the way the FTSE 350 Higher Yield index is definitely not HYP, but it is at least investing in the same pool of higher-yield mid-to-high cap UK shares as my HYP strategy is (**); given that choice, the obvious candidate to compare it with is the FTSE 350 Lower Yield index. So if I calculate ((end date FTSE350HY)*(start date FTSE350LY)) / ((start date FTSE350HY)*(end date FTSE350LY)) for as many pairs of three-year-apart start and end dates as I can, using the total-return versions of the indices, and look at the proportions of results above and below 1, I'll get some indication whether a period of 3 years is enough to come to any serious conclusions about whether high-yield strategies are better than the alternative, at least in that pool.
I've just done a reasonably close approximation (***) to that. The results were that 1793 out of the 4092 3-year periods covered by my data (43.8%) gave a result above 1, indicating that high yield had outperformed low yield over that 3-year period, and 2299 of them (56.2%) gave a result below 1, indicating that high yield had underperformed low yield over that 3-year period. So any conclusion drawn from a
single 3-year comparison within the 19 years and a few months covered by the data I'm using has at least a 43.8% chance of being wrong about which strategy choice is better - which IMHO means that any idea that such a conclusion is a
serious conclusion should be laughed out of court... I wouldn't describe it even as a tentative conclusion - it's just a very preliminary indication.
One might ask whether a few more years would make a significant difference to that. So I've tried it for 5-year periods rather than 3-year periods. The result was that 1058 out of 3588 periods (29.5%) said high yield outperformed and 2530 (70.5%) that it underperformed. Better, but nearly a 30% chance of giving the wrong answer means IMHO that it's still a preliminary indication rather than any sort of conclusion.
How about 10 years? Well, doing the same exercise for 10-year periods has 692 out of 2328 periods (29.7%) saying that high yield outperforms and 1636 (70.3%) that it underperforms. So apparently going up to 10 years produces little change to one's chances of getting a wrong answer. But I don't think that can be treated even as a serious estimate of the chances of a single 10-year period giving the wrong answer, because all the 10-year periods it looks at have something in common, namely several months in 2009 - this approach studies periods that
all have something in common with each other if they are longer than half the length of the data you have available. And indeed, they
mostly have something in common with each other if they're close to half that length - an effect that means I wouldn't put much credence in the approach's estimates of the chances of getting a wrong answer for periods over 5 years, and that may well noticeably reduce their reliability even for 5-year periods. About all I can say is that if e.g. I had another 20 years of data, allowing me to look at 5040 more 10-year periods, at least 692 out of the 7368 periods (9.4%) would have high yield outperforming and at least 1636 of them (22.2%) would have high yield underperforming. So I would still be getting the wrong answer out of at least nearly 10% of 10-year periods - and it wold only be that low if
every additional 10-year period added had high yield underperforming - which seems unlikely to me!
Finally, please note that the above is about whether looking at a single period of 3 (or 5 or 10) years will allow you to draw serious conclusions about whether one particular high yield strategy (which has features in common with HYP strategies, but is not actually a HYP strategy) outperforms its low-yield equivalent, and the answer it comes up with is "No, it isn't enough for serious conclusions, not by a long way!". It is
not about what the answer to the question of whether that strategy does actually outperform its low-yield equivalent actually is, just about whether 3 years of experience of that strategy is enough to be at all confident that one knows the answer. And just in case anyone thinks that while any single 3-year period isn't enough, the 56.2%-to-43.8% 'vote' of all the 3-year periods in favour of low yield is convincing, the high-yield strategy of the FTSE 350 Higher Yield index actually outperformed that of the FTSE 350 Lower Yield index by 67.7% over the entire 19-years-and-a-few-months data range I've used, due to a strong tendency to outperform by more when it outperformed than it underperformed by when it underperformed.
It doesn't actually say anything about whether 3 years is enough to draw serious conclusions about whether a particular HYP strategy out- or underperforms its low-yield equivalent, of course, but one can do quite a few similar exercises for other strategy choices, e.g. the large-caps-vs-mid-caps choice with the FTSE 100 index against the FTSE 250 index. That one, for which I've got a bit over 25 years of data) comes out as the FTSE 100 outperforming the FTSE 250 in 23.5% of 3-year periods and underperforming in 76.5% of them, outperforming in 16.0% of 5-year periods and underperforming in 84.0% of them, outperforming in 0.0% of 10-year periods (and that is actually not a single one of them, not just so few that the percentage rounds to 0.0%) and underperforming in 100.0%, and outperforming by 50.6% over the entire data range. So that one is pretty clear: mid caps outperform large caps - but 3 years is not enough to get more than a preliminary indication of it, 5 years is probably just edging into "tentative conclusion" territory and 10 years is the sort of period required to arrive at a serious conclusion.
As yet another example, how about the decision whether to invest in equities at all? Well, a similar exercise for whether investing in the FTSE 100 outperforms a cash-under-the-mattress strategy (i.e. not even investing to earn interest) has the FTSE 100 outperforming cash-under-the-mattress in 69.3% of 3-year periods, 76.3% of 5-year periods, 93.9% of 10-year periods, and by 473.2% over the entire ~25-year data range. So three years of personal experience isn't even long enough to reach a serious conclusion about whether to invest in equities at all... I could go on, but I think that's enough to make it clear that Ian is basically right to say that 3 years is too short a period to produce any serious conclusions about investment strategies. Not absolutely 100% right, but everything I've done along those lines says that the conclusions that one can seriously come to on 3 years of personal experience are the ones that are pretty obvious anyway, such as that investing cash by putting it on deposit with large, well-capitalised banks and earning interest on it outperforms cash-under-the-mattress!
(*) Usually expressed as the percentage (factor - 1) * 100% - e.g. a performance factor of 1.16 is normally expressed as a performance of 16% - and that transformation can be reversed to give factor = (percentage / 100) + 1. But the maths here is simpler and easier if one works with the factors and only does the conversion to percentages right at the end. And by the way, I'm only writing a lot of this for any readers who are not familiar with this way of doing things - I know I'll be teaching many grandmothers to suck eggs, but that's unavoidable when one wants to cater for an audience that is likely to include both grandmothers and non-grandmothers.
(**) Someone whose HYP is fishing in a different pool - e.g. by going outside the UK - might want to look for a different proxy.
(***) The approximation is that (a) I've only used the data I have handy, which is the index values for each trading day from 18/01/1999 to 06/04/2018, missing 9 scattered days that I believe were trading days; (b) the periods are from a line in my data to the 756th line after it - at 252 trading days per year (5 in each of 52 weeks minus 8 bank holidays), that is 3 years' worth, but the period they cover can be a few days different due to years being 1 or 2 days longer than 52 weeks, timing of weekends and bank holidays, and the missing trading days. The 5-year and 10-year figures above are similarly actually for periods of 1260 and 2520 trading days, excluding the few trading days I'm missing data for.
By the way, I obviously could have brought my data up to date, extending it by nearly 15 months. But that's not enough to be able to make a serious difference to the above, and it's a bit (through not greatly) more effort than I was prepared to put into this post.
Gengulphus