The "real return" investment strategy

I've written this description of my investment strategy for early retirement mainly to flesh out, clarify and summarise my thinking. Although I address an imaginary audience, I'm mainly talking to myself. Anyone who reads this should make up their own minds about its merits, and take responsibility for their own decisions if they allow themselves to be influenced by it.

1. Investment phases and early retirement

   My "real return" strategy was intended to address the issue of how to safely achieve the maximum possible income from investment assets over a long retirement period.

   In theory the fact that your early retirement savings only have to last as long as you live means you can count on withdrawing an element of capital each year. In practise the uncertainties of a long retirement period mean that I feel it is safer to divide your retirement into two phases, a coasting period where you try to maintain the value of your capital, and a run-down period when your life-expectancy is short enough that it is safe to start spending capital. In practise I'm unlikely to be certain that I want to use all my capital up on generating an income until my life expectancy has fallen to about 15 years, and if I do decide at that point there's nothing else I want to do with it, my income will be much higher in those years than the earlier ones. This isn't ideal. If spending throughout retirement is constant then ideally income should be as well. The marginal utility of money means that extra income in one year would have been worth more to me if it were spread equally between all the years. Unfortunately it seems unavoidable that being ready for contingencies does mean spending less than the maximum possible income, for the majority of a long retirement.

   The "real return" investment strategy involves calculating how much income you can take during the coasting phase without your capital being eroded by withdrawals, charges, taxes or inflation. The real return tells you how much capital you need in the first place. The returns on investment are not treated as adjustable. It is outputs (spending) or inputs (initial capital) that are expected to adjust to the returns that are available.

   In fact the real return strategy is a sensible strategy at any stage of life. It can be used in the accumulation phase, when one is still generating capital from working. A variation of it could also be used in the run-down phase, with the income taken being higher because the capital is no longer being preserved. In practise, with fifteen years to go, I'm most likely to just buy an increasing or inflation-linked annuity.

   Once you've calculated the real return on different asset classes, the result also affects your asset allocation.

   Before I get to the calculation, I need to visit some ideas that affect the strategy. The first couple will be familiar to anyone with any knowledge of investing, but some more interesting ones follow thereafter.

2. Diversification, asset classes and asset allocation

   The idea of diversification is to hold different investments, so that when some do badly this is offset by others that do well. In the case of shares, splitting capital between as few as 15 companies that are all very different from each other is the bare minimum required for diversification. In practise, it is easier to invest in a fund, which will invest in a far wider spread of companies. Choosing a share tracker fund is an easy way to ensure we are as diversified as we can be, within the market the fund tracks.

   An asset class is a group of investments that tend to be affected by the same factors, and therefore move up and down in value together. It is slightly subjective what we decide to label as asset classes, as there are always some factors that will cause any two investments to move in the same direction when they change. Some people divide shares into different asset classes, such as by country or region, or company size. Sometime a market sector is treated as an asset class. As shares of all kinds tend to move in the same direction during big stock-market booms and crashes, I prefer to regard shares as just one big asset class.

   The other asset classes I recognise and use as a UK investor are cash deposits, bond funds and commercial property funds. Hedge funds and commodities are two asset classes I recognise, but don't regard as useful. I'm excluding residential property from consideration as an asset class on the assumption that the early retiree owns their own home outright, which in the UK in 2007 probably means having more money in one place than is strictly speaking a good idea.

   The commercial property funds I have in mind are the unitised funds offered by insurance companies within pension and other investment products that they offer. Property unit trusts have unacceptably high charges. Property investment trusts do not have the very low volatility that is one of the best features of the asset class. They also have much higher charges than the cheapest funds available from insurance companies. Lastly, they use gearing, which I see as a negative factor. At the time of writing share prices of property investment trusts have just crashed by 40%, and they may therefore be a better temporary home for money invested in commercial property than unitised funds, but in the long term I will return to insurance company unitised pension funds for my property holdings.

   One feature of unitised property funds that operate single-pricing is worth noting: when the managers switch the unit price between offer and bid bases, the fund value can fluctuate by 6% overnight. (Single pricing isn't such a big issue in share and bond funds, because the difference between the buying and selling prices of the underlying assets is much smaller.) Fortunately such switches are rare. I've noticed that Standard Life have adopted what I think is a clever strategy to stop people trying to trade against the switch. When their life fund switched to a bid basis for the first time in its 28 year existence, they closed it to inflows and started a new fund priced on an offer basis for anyone wanting to switch into property.

   For the same reasons we diversify within asset classes, we need to consider diversifying among asset classes.

   There is a formula based on finance theory for calculating the optimum mix of assets one should use. One problem with this is that it relies on parameters the estimates of which are so uncertain that the actual optimum portfolio could easily be very different from the one we calculate. Also, I read somewhere that "naive" asset allocation, consisting of simply allocating equal amounts to each asset class, is usually not much worse than the optimum allocation. I prefer naive allocation. It is easy to do, and means you will often have a big chunk of money in reserve to move into a different class after the latter has crashed. (The optimum mix formula will usually allocate a lot of money to one class and much smaller amounts to others.)

3. Market efficiency

   Markets are efficient if whenever you come to trade the share price has already adjusted to all information that should affect it. This means that the share price is never expensive or cheap. The idea is quite logical; if there were any information anywhere in the world that would lead sensible people to believe a share price was wrong, they would buy or sell the share, and the price would adjust, and the process would continue until the price was right. As there are always thousands of highly-motivated people studying each share and buying or selling, the price should always be right.

A consequence of an efficient market is that there is no point in actively trading shares or paying someone to choose them for you. In an efficient market, any two diversified portfolios of shares will, with average luck, return the same as each other. By trading actively you increase your costs without increasing your returns. Similarly, anything you pay anyone who chooses shares or funds for you can only reduce your returns, as you will be paying a cost and receiving no benefit in terms of increased return. You must pay as little as possible to fund managers, financial advisors or full-service stockbrokers. It's important to remember the phrase "with average luck." Of course different portfolios may return different amounts. The point is that in an efficient market the differences are due to luck, not expertise. There's no way of knowing in advance which fund will return more, and therefore no reason to pay any fund manager more than the amount charged by the cheapest one.

Another consequence of market efficiency is that it is not worth spending time analysing or researching shares, since any information you come across will already be reflected in the share price when you come to trade.

There's an apparent paradox implicit in the idea of efficient markets. Markets can only be efficient if a substantial number of participants believe they aren't. If everyone believes markets are efficient, no-one will research shares, the share-prices will not correctly reflect the prospects of companies, and markets will become inefficient. If markets are inefficient, it pays to research them, so people will, and they will become efficient.

The investment classic that popularised the idea that markets are efficient is "A Random Walk Down Wall Street" by Burton Malkiel. I believe that when this book was first published, tracker funds didn't exist, and that they were invented to satisfy a demand created by this book. A cheap tracker fund is the ideal investment for a believer in efficient markets.

4. Market rationality

   I read "A Random Walk Down Wall Street" for the second time shortly after the stock-market crash of 1987. Although 95% convinced by it that I should treat stock-markets as being efficient, I found its explanation of how the 1987 crash could be consistent with efficient markets theory difficult to swallow.

   I define the term irrational market to mean one where share prices do not correctly reflect the prospects of companies, but it is impossible to predict when they will correct themselves. I replace the idea that markets are always efficient with the belief that they are always some mixture of efficient and irrational. In an irrational market, most of the practical rules of thumb that guide a believer in efficient markets still apply. There is still no way to get a higher return by predicting the future, whether directly by doing research, or indirectly by paying someone to choose shares for you.

   It's something of a paradox that you can be absolutely certain that the stock-market is overvalued, and even though you are right, there is no sensible way to profit from this insight. You might consider buying a put option, however options have a time-cost built into their price, and (to quote a famous phrase attributed to Keynes) "the market can remain irrational much longer than you can remain solvent." If you try to profit by shorting shares, note that an overvalued market can become twice as overvalued, possibly bankrupting you, before the expected crash comes. On the other hand, you can use your belief that a market is overvalued to avoid investing in that market, and this is one strategy that differentiates someone who believes markets are sometimes irrational from someone who believes they are always efficient.

5. Sequence of returns

   Uncertainty about the actual sequence of returns you might experience makes how much you can regularly take as income without prematurely running down your capital a much more difficult issue than it first appears. It's not good enough to take no more than the average amount your investments are expected to return. It makes a huge difference what order the good and bad years arrive in. Even if you get the expected average return, if you are unlucky and the worse-than-average years are disproportionately grouped at the beginning of the withdrawal period, you will eat into your capital and your portfolio will never recover from the dip. It is possible to take income at no more then the average annual rate your investments actually deliver, measured from your retirement date, and not only fail to preserve your capital, but actually spend it all. Some authors suggest on the basis of historical data that the maximum income a typical portfolio can safely support over a long period is 4% of the initial capital. A long period typically means 30 years or more. You are allowed to increase the income each year by the rate of inflation, so that it stays the same in real terms. Note that by safe they just mean you are highly unlikely to run your capital down to zero before the end of a 30 year period; your capital may still fall to near zero by the end.

6. Market valuation

   During the late 1990s I strongly suspected the stock-market was overvalued. I'm talking about mainstream shares here; the dot-com bubble never came close to making any sense to me. Eventually I found intellectual justification for my instinct to stay out of shares. In March 2000, before the mainstream market crashed, "Valuing Wall Street" by Andrew Smithers and Stephen Wright was published. They said that stock-markets had a fair value that you could calculate, and this might differ from the current market value given by share prices. They advocated something called the "q" ratio as a measure of value. There is a second measure that also works, and gives the same answers, if calculated correctly. This is the cyclically-adjusted PE ratio of the market as a whole, called CAPE for short. Overvaluation of both is measured by comparing them to their historical averages. "q" is theoretically better, since there is a fundamental economic reason why it must remain constant on average over the long term. In the case of CAPE, its long-term average will only remain constant if the average return on shares remains constant. This return, known as Siegel's constant, looks as though it has stayed constant at about 6.75%, but there is no economic theory that explains why it should be at this level. As a result, it is always possible for market-valuation sceptics to argue that the long-term average value for CAPE might be changing over time. Having said that, Smithers and Wright prove that if CAPE is correctly calculated, it must give the same result as q. The charts in their book show that most of the time it does indicate the same level for the fair value of markets as q does. At the peak of the year 2000 stock-market bubble, the log of the ratio of q and CAPE to their historical averages peaked at about 1, a value that means the SP500 index was at 2.7 times fair value. The second highest peak was just before the 1929 crash. According to a chart at www.smithers.co.uk, the ratio then was 0.7, which means the index was at 2.0 times fair value. The biggest undervaluation was in about 1920, when the index was somewhere between 30% and 40% of fair value. The range of undervaluation I quote is because this was one of the infrequent periods when there was a noticeable divergence between the two measures. At the time of writing there has been another divergence. In June 2007, the date the chart was made up to, the SP500 was at 1520, q was saying the SP500 index was 29% overvalued and CAPE was saying it was 83% to expensive.

   Smithers and Wright showed that in the long-term, by which they meant periods as long as 30 years, over or undervalued markets would correct themselves. This long time-scale is important, since it means their theory does not entirely contradict the idea that markets are unpredictable. Efficiency and irrationality still mean it is not possible to predict where markets will go over much shorter periods.

   In early 2000 the high price of shares meant the long-term return was so low that taking all your money out of shares was the only sensible option.

   Using the figures for June 2007 as an example, I can calculate the future return I expect on shares. First, I get the level of overvaluation indicated by q from www.smithers.co.uk. (I could take the equivalent figure for CAPE, or when they diverge, the average of the two, but I've decided that I trust q on its own the most.) The level of overvaluation is currently explicitly mentioned, but if it weren't I could derive it as follows. I get the latest value of the ratio of log q to its historical average, 0.25. (Actually for this example I cheated and calculated this from the stated level of overvaluation, if I'd read the value off the rather tiny chart I might have guessed it was closer to 0.20.) In Microsoft Excel I calculate the fair value of the SP500 index as 1520 / exp(0.25) = 1178, where 1520 is the level of the index at the end of June. The actual level of the index divided by fair value tells me how overvalued it is, 1520 / 1178 = 129%. 100% indicates fair value, so it is 29% above fair value. I divide Siegel's constant by this ratio to get an adjusted return. This gives me an expected return 6.75% / 129% = 5.2% for shares. (If I'm making the calculation a few months later, I can adjust the fair value by assuming it should have risen by (5.2%-1.5%) per year, where 1.5% is the dividend yield of the SP500.) Since I'm a UK investor, I then simply assume that the yield on the SP500 index is the best estimate of what I can expect from whatever UK fund I do invest in. This is a big assumption, but I don't have any alternative, as comparable data isn't available for other indices. In practise I have charted the SP500 and UK indices together, and there is a high degree of correlation. At the very least, I expect major bubbles and crashes to occur simultaneously in both, and if my method steers me in the right direction with respect to those, that's a huge advantage compared to investing in the dark.

   While writing this, I've become aware of an academic paper on the web that not only agrees that markets can be valued using q and CAPE, but uses regression equations derived from historical data to predict the market over different time periods. The regression equations for shorter periods currently give far more pessimistic predictions than my method. The paper itself only contains regression equations for periods up to 20 years, however the site that referred me to it has its own version of the regression equations, and for periods of 30 years or more they give similar answers to my method. The reason the regression equations predict worse returns over the short term is that the historical data appears to show that there is momentum and feedback in the market, which in a bear market means bad years beget bad years. The equations collectively imply that when the market is overvalued, any correction is not likely to be spread evenly over a large number of future years, most of it is most like to come in the earlier years. The implication of this is that even if the long-term return is currently something like the 5.4% I've calculated for December 2007, and even if that were the highest return available on any asset class, I should still stay out of shares because the return over the next several years will probably be highly negative. A much better time to switch into shares will arrive, if I wait. Links to the paper and the site that referred me to it are at the bottom of this page.

   Having made my own attempt to develop regression equations on the basis of historical data, I'm not convinced they are useful. Using recent stock-market history (in addition to valuations) did make a slight overall improvement to prediction accuracy, but for some reason that I failed to get to the bottom of, using these shorter-term predictions in a test of a real return strategy based on historical data actually seemed to deliver worse results than using long-term predictions based on valuations alone. I've decided to stick to my method outlined earlier in this section.

7. The real return approach to investing.

   1. Decide how much income you need to generate. You need to leave room for things to go wrong, including the possibility that the following plan is flawed. My target figure is 133% of what I need to live comfortably, and double the minimum I need to pay my bills. I assume I'll usually only draw three-quarters of my target income each year, leaving the rest for contingencies.

      There is a second good reason, in addition to allowing for contingincies, for taking an income well below the maximum. Happiness, to the extent it is determined by income, depends on keeping pace with the spending of your peer group. If income grows with inflation, and we use an accurate inflation rate, that only ensures we can continue to afford the things we know at the outset we are going to want to spend money on. It may not cover new things. A person retiring in Britain 30 years ago would not have budgeted for things we take for granted today, such as satellite television subscriptions, internet access and annual foreign holidays. Ideally, your income needs to rise in line with wage earnings, which generally increase at a higher rate than inflation.

   2. Regard your capital as divided into pools. Money is in the same pool as other money if it can freely be moved to and back from the same place as the other money. For me, pools are defined by the tax system. Pension investments are one pool, tax-free individual savings accounts (ISAs) are another pool, investments held directly in non-tax-privileged accounts are a third. To give an example, money held in a fund in a Legal & General Stakeholder pension is in the same pool as money held directly in shares a Sippdeal account, because it is a legal requirement that pension funds can be transferred from one provider to another, on demand. On the other hand, investments held in an ISA are not in the same pool as investments held directly outside an ISA, because it is not possible to move more than £7000 per year into ISA accounts, and one would not want to move money out of an ISA unnecessarily, as it would permanently lose its tax-free status. It is however possible to move investments between ISA accounts with different providers without penalty, so all ISA investments are in the same pool as each other.
   3. Estimate how much income different asset classes will support indefinitely.
      1. For cash subtract the rate of inflation from the expected interest yield to get the income yield. Treat any borrowings such as a mortgage as a negative cash holding that produces negative cash income.
      2. For bond funds, subtract the rate of inflation from the redemption yield to get the income yield.
      3. For shares, use the yield calculated as described in the section on market valuation.
      4. For property funds, use the net rental yield.

      Note that inflation is only subtracted in the case of cash and bonds. I broadly expect the capital value of "real assets" such as shares and property to keep pace with inflation. The "real return" I'm calculating is the amount that can be taken as income without the capital being depleted.

      Once you've calculated the overall return you expect on each asset class, calculate modified results separately for each pool, lowering these theoretical yields to take account of any taxes and fund management charges that will have to be paid on the actual investments in that pool.

      Here are examples of how I estimate the returns on different asset classes, in my pension pool, in December 2007. I could get a yield after inflation and charges of 5.5%-2.1%-0.4% = 3.0% from Legal and General's cash fund. The yield after inflation and charges for a bond ETF held in a Sippdeal account would be 6.4%-2.1%-0.2% = 4.1%. The yield of 5.2% in June 2007 for shares becomes 5.4% in December, when I adjust for changes since June. This is reduced to 5.0% by tracker fund charges. Data compiled by IPD and published from time to time in the newspapers gives the industry-average rental yield on commercial property, most recently 4.6%. As unitised funds have declined by about 15% since this was published, I adjust this by 15% to 5.3%. This gives the figure I would use for the unitised property fund that is available in a Legal and General Stakeholder pension, where tiered charging means that I would pay 0.4% if all my pension funds were held with them. This charge would reduce the yield on unitised property to 4.9%. However December 2007 falls in a highly unusual period for property, as property investment trusts are trading at huge discounts of between 30% and 50% to asset values. As a result, I am temporarily favouring investment trusts held in a Sippdeal account for property. The figure I use for the real return I expect from this variant of the property asset class is the rental yield, net of running and finance costs, which I calculate from the companies' latest accounts. The median yield for the funds I follow is currently 6.7%, compared to a median dividend yield of 8.8%. I'm not happy to take the dividend yield as the measure of income, as I can't know for certain that allowance has been made to prevent inflation and management charges eroding capital.

   4. Within each pool, decide which asset classes you are going to use, then split money evenly between them. When first implementing the strategy, use all asset classes available within the pool that give a return within 0.25% of the most promising one. For example, when predicted returns are cash 3.0%, bonds 4.1%, property 6.7% and shares 5.0%, money would be invested entirely in property. As asset classes rise and fall from time to time you may have to switch completely into or out of any particular asset class. To prevent to frequent switching, add an unused asset class to the current mix only when it has the highest yield, but only eliminate one from the current mix when its yield falls more than 0.5% below the highest. Where a pool is invested in more than one asset class, buy whatever yields the most when adding funds or sell whatever yields the least when withdrawing money from the pool.
   5. An alternative strategy would be to always split money evenly between the best two asset classes available in each pool. Frequent switching could be prevented by only allowing changes once a year. Maintain the even balance by shifting funds once a year, or by buying whatever you have least of when adding funds to the pool, or selling whatever you have most of when withdrawing money from the pool.
   6. Take into account the costs of switching, especially capital gains tax where applicable, by having a rule that only allows a switch if the yield in cash (rather than percentage) terms from the whole portfolio would be higher after the switch, even though the capital may have reduced, for example by money set aside to pay any CGT liability the switch has caused.
   7. The amount of capital you need to retire is the amount needed to generate the income you specified in the first step, given that the capital is invested and income calculated as described above.

   After retirement, the maximum income you can take can be calculated at any time in the manner described above. In theory, if the yields you used for each asset class were always accurate estimates and never had to be adjusted downwards for any reason other than a rise in capital value, the income the portfolio produces would never fall. The income should rise slightly from year to year in line with the long-term rate of increase of company profits and property rental income. From time to time there could be a sudden permanent upwards jump in income, when an asset class crashes to become the new highest yielder, and you switch funds into it. Most of the time it wouldn't matter what happened to the value of your capital, as any reduction in capital values would be offset by an equivalent increase in the yield, leaving the amount of income you can take from the asset class unchanged. (This rosy scenario of never-falling income is only a broad-brush picture, in practise there are many things other than yield estimate miscalculations that could cause a drop in income yields without an offsetting rise in capital. The changing yield on cash is never offset by corresponding changes in capital. In a long recession share earnings and property rental yields might decline. The yield on bond or property funds might decline as a result of unfortunate trading by managers. Bond issuers may default at a greater rate than was anticipated in bond fund prices at the time of purchase.)

   My real return strategy would wisely have required you to have far more capital if you were contemplating retiring in 2000 than it would have if you were doing so a few years later, after the stock-market decline.

   I have looked at the effect of applying a variation of this strategy with just two asset classes, shares and "cash" (the long interest rate in Prof. Schillers data) over the period 1881-2007. Using hindsight, I tuned the actual withdrawal rate to be 87% of the predicted rate, the predicted rate being the forward-looking yield for the chosen asset class at any given time. This tuned withdrawal rate ensured capital at the end of the period exactly matched that at the start. In this strategy I assumed a 100% switch into the more promising asset class immediately it took the lead, with switching potentially on a monthly basis, and no switching costs. The adaptability of "real return" strategies is illustrated by the varying withdrawal rates I found. In July 1921, when the stock-market was at 41% of fair value, and inflation was -14.9%, this strategy suggested an annualised withdrawal rate of 17.8%. In October 2000, a couple of months after the market peaked at 294% of fair value, it suggested a withdrawal rate of just 2.03%. In both cases knowledge of what happened next makes these extreme withdrawal rates look justified.

   I doubt I'd ever be brave enough to take the monthly equivalent of 17.8%. It's worth mentioning that the strategy very seldom recommended double-digit withdrawal rates. The median annualised withdrawal rate was 6.0%, and the average was 6.7%. It has not indicated a rate of 6% or higher since February 1991.

8. Summary

   1. This method incorporates rules suggested by efficient markets theory by only considering asset classes as a whole in choosing where to invest. Within each asset class, efficient markets theory tells us we should simply choose the cheapest diversified fund.
   2. It helps deal with market irrationality by taking into account market valuation. It is likely to take you out of an asset class during an extreme boom, and into it after an extreme crash.
   3. It guides you on asset allocation by making you look at the return you can expect, and tells you how to do this in the difficult case of shares. It maximises returns by only spreading your investments between different asset classes when the differences in expected returns are not to large. The alternative allocation strategy of always using the best two asset classes gives you a more cautious option.
   4. It helps deal with the sequence of returns problem during withdrawal by usually suggesting low incomes that are not to far from generally recommended safe levels, but improves on those by tuning the result to the current state of markets. In theory, it only allows you to withdraw the real income generated by the underlying assets, leaving the capital untouched. (The fact that market fluctuations cause the capital to vary in value shouldn't matter, as long as it continues to generate the same underlying income stream.)
   5. It does not completely deal with the sequences of returns problem, or the possibility that real returns might fall after retirement. It is necessary to monitor and maintain your strategy, and possibly make adjustments to your spending if your retirement is going off-track.

9. Links

   1. If you have any doubts that stockmarkets can be valued and long-term returns predicted, Valuing Wall Street is your new bible.
   2. This paper includes regression equations to predict the SP500 index in 1, 5, 10 and 20 years time, based on current valuation levels. Note that the equations written on the charts for predicting one and ten year returns are wrong, you can derive the correct ones from the table "Exhibit 8".
   3. The site Early Retirement Planning Insights gave me the link to that paper, and contains calculators based on its owners own versions of the regression equations
   4. Somewhere on the web site of Smithers & Co you will find data or charts indicating current levels of valuation.
   5. Firecalc
   explains what I have called the sequences of returns problem much better than I have, and contains an interesting calculator for calculating safe withdrawal rates.