I have been feeling slightly depressed this last week or two. This is because of a growing feeling that I won’t be able to quit my day job as soon as I would like. Well as soon as I would like would be yesterday but you get what I mean.
I did a simple average of my alphas for all 15 of my ports (actively traded and edited). Clearly this is not really valid if for no other reason than some of my ports have 5 stocks and others have 10. But I did it anyway. My average alpha is 21%. While I would have been very pleased with this a year ago (before P123), my book simulation (backtest) has an alpha of 50%. So am I doing something wrong? Can I do better. Of course, the answer is you can always learn and do better but what can rationally be expected?
So what are the statistics for R2G ports’ annual excess? The average annual excess for all of the ports since launch is 11%. Since inception, it is 44% Hmmm. What does this all mean. I don’t know.
Any ideas on better calculations, expected difference between inception and launch, what a good alpha (or other statistic) is, realistic expectations of alpha etc. would be greatly appreciated.
More to the point, any ideas on how to quit my day job sooner?
Slow and steady wins the race. Don’t be so dazzled by the mirage that you forget to pick up the coins in front of you.
Buffett is the worlds richest man and his annualised return is 21% (?) - the mathematics of exponential growth are astonishing. The mathematics of risk is, on the other hand, the killer. A 50% loss requires a 100% gain. A 75% loss requires a 300% gain. A 100% loss…
I can’t tell you how you can quit your day job, but i can comment on some of your other points.
The most important thing to note is that any alpha that’s more often than not above zero is excellent. Yes, I really said zero. That is the realistic threshold.
Remember what alpha means; it’s the measure of return attributable to the actions of a portfolio manager (separate from the market) adjusting for the fact that the manager may have assumed above- or below-market risk. The efficient markets theorists say alpha is impossible. Many in the real world, including at p123, have shown otherwise. What’s important is that you remember the threshold of excellence – anything above zero more often than not.
Now, the gap between annual excess since inception versus since launch is quite easy to explain, and if you follow the boards, you’ve probably seen much that has led to this. Backtesting/simulation is a tool. a spectacularly powerful and valuable tool. But like all valuable things, it can be misused. (Consider the automobile, a fabulous transformative tool for humanity but one that does occasionally get misused resulting in all sorts of mayhem.) An example of misuse is when the backtest/sim result becomes the goal, rather than what it’s supposed to be; a feedback mechanism to confirm or challenge the validity of your ideas. Optimizing – picking factors that produce the best possible backtests often with little understanding or concern with how such factors should impact live performance is a path to pretty much what you point out; disappointing live performance. Also, when backtesting, it should be done thoughtfully with regard to market realities. Don’t just pick the MAX backtest. Understand what’s been going on in the sub-periods and which sub-periods are most likely to constitute a reasonable sample. There’s also been much discussion here around liquidity and slippage. It’s very very easy to juice backtest results by going way down the liquidity scale. The worst penalty you “pay” is a slippage assumption. But there is no slippage assumption that can come close to representing what can happen in the real world to those who don’t behave sensibly. If a big quantity tries to move too fast, I’ve seen “slippage” of 20%-plus, once not long ago I saw 65%. So don’t try to spike tests with low liquidity; it’s a path to nowhere. (If you really have a good low liquidity model; trade it personally – that’s where you’ll make your money. That’s what I do. R2G, like any category of professional models, needs liquidity.)
Try to build wealth slowly, let compounding be your friend, and be mindful of Marc’s advice here too … i.e., be rigorous about your testing, incorporate conservative assumptions about liquidity, slippage, etc., and focus on the journey, not a quick outcome.
On a more practical advice side. If you are near retirement and just looking for a threshold savings then maybe you quit when you reach it.
However, if you are mid-career hoping to quit your primary income and live off of trading based on your trading experience during what has recently been a very good market (you only joined P123 a year ago) you need to think really long and hard about that. One year of out of sample duing a good run in the market isn’t much to base giving up a current career on unless you have saved a ton. Make sure emotion isn’t clouding your judgment like job disatisfaction, boredom, wanting something different and you aren’t overestimating returns. If anything if your real Alpha is 11% in your out of sample, you should be counting on something less than that to base quitting a career on or you are looking for disaster. If you have minor children then you should be even more conservative in your estimations IMHO.
What time period is your real time alpha? If it’s a short time period then it’s just meaningless noise. I’m relatively new here. I’m trading 2 R2G’s. Both gave me negative alpha starting out. Now one has positive alpha and the other is still negative. I plan to trade them at least 3 years before I can gather any meaningful data. So these short term numbers don’t mean much.
I fully expect real time alpha to be lower than backtest alpha. How much is impossible to tell. Another thread mentioned that cutting the alpha in half is a good rule of thumb. The fact is, the backtest is based on historic data, and the future will always be at least a little bit different. I guess it’s possible that real-time performance is actually better than the backtest, but that’s just pure dumb luck.
Forecasting your retirement date is hard. A few percetage points difference in your annual returns can add or subtract quite a few years, due to compounding. There’s alpha that your model generates, but there then is the benchmark, and if markets and the economy does really well for a long period, you can retire early. If we have an extended period of recession/depression, then there’s not much you can do about it.
I’m like you, I have a job and dream about financial freedom. A job gives you a stable income but won’t make you rich. To get rich you have to do stocks, or real estate, or start a company. All involve risk.
@Chang. Very short time periods. I posted previously that I run 3 different weekly ports but do it each day of the week for a total of 15 ports. So I averaged the Alphas of each of the 15 ports. This is an attempt to get more information from what I agree is a very short period. So perhaps better expressed as the average of the 3 different ports (5 ports for each average): 27, 23 and 12. The 12 is the mean of the 5 stock ports and will probably have more noise than the 10 stock ports as you say. Also, perhaps, should have had half the weight.
I’m beginning to experience first hand that what you say is true: [quote]
I fully expect real time alpha to be lower than backtest alpha. How much is impossible to tell. Another thread mentioned that cutting the alpha in half is a good rule of thumb.
[/quote]
Hats off to those who have authored R2G ports that have performed well out-of-sample.
In trying to measure Alpha over the short term, in a bull market, you might just be getting beta. I’ve found out the hard way that strategies that can make a lot of money quickly also have the potential to lose a lot of money quickly. If you want to make an income from trading I suggest that you save aggressively and diversify your trading so that you can make (or at least not lose) money in most any environment.
Jim,
below are my views on how to measure what is true out of sample performance in actual trade results.
It is a combination of High Annualized gain (Alpha), Low Draw Down, Lower Annualized Turn Over(less slippage), Higher daily liquidity (less slippage), High Average gain%/no of days hold, lower number of trades per year plus any other factors that I missed…other could insert.
Example :
Model…Ann Gain…DD…Ann Turnover…Daily Liquidity…Avg Gain%/no days hold…Total no trade/year
A…+90%…-25%…2000%…>$500k…5%…15days…400
B…+90%…-25%…1000%…>$500k…8%…28days…280
If we compare above two hypothetical cases, which one can offer better actual real life out of sample trade results ?
I think it should be model B with less trades and higher gain% per trade and potentially suffer less from slippage as a result of less turnover though both models have the same daily liquidity.
You need to examine which case your actual situation is closer to, A or B.
Jim, consider relocating to a rural area far away from any glamorous city. Real-estate will be 20% to 90% cheaper, goods and services at least 20% cheaper. If you are really desperate to retire you can move to a country like Thailand, or others in central/south America or Asia. In many countries you can live well with $1000/month or less. Thailand for ex offer visa for people who wish to spend their retirements there. If you want extra cash you can partner up with a local person and open a burger joint or something. I’m not kidding. I’ve just relocated to a rural part of France, bought a 900 sq feet house for the price of a parking spot in Paris (true story !). Everything is farther away in distance but since there is no traffic jam it takes the same time or less to drive to the airport/train station etc… and internet is faster (go figure).
About quitting your day job: you can get farther for the same money and the same (or better) quality of life in some low-cost, low-tax, high-temperature countries. Tips: choose a place with a history of low inflation and low real estate prices, do about one year of feasability study from home, rent at least 3 years before buying or building a house, and keep always enough cash to fund a few years of decent lifestyle and a comeback. I did the jump a few years ago. It may be more complicated if you have children going to school.
So you can live welll on a 100 thousands euros portfolio…besides you con publish some R2G portfolios and earn an extra money…
Say you can get a 20% on that port…20 gran on a rural area can be good.
The thing is…What cagr can you reallistically get with R2G portfolio on a 100.000 euros?
Thank you!
I’d love to move. Just for the adventure of it. But I’ve got two little girls in school. I write and trade. I don’t need to be in LA for that. And taxes and real estate here are nuts. But, I love the weather and local community. And love my kids schools.
How’s Paraguay? Where’d you move from? What other places made your final lists?
@quixote1: don’t rely on a single strategy or even on a single tool (even one of the best like P123). My humble opinion is that being an independent investor is having enough strategies with distinct bias/rationales that may possibly be run without a software. Otra cosa: hay que hacer algo fuera de tus inversiones. Mirar tu dinero en una pantalla todo el dia no es una vida
Once you retire, I don’t think it’s safe to assume anything close to 20%/yr. returns. Assuming no real salary. With no salary, you’re risk profile and reaction to DD’s will likely change significantly. And markets can underperform for a very long time.
You need a) 2 years living expenses in cash, b) realistic inflation assumptions that leave a margin for safety, c) a good understanding of risk, d) a diverse combination of models and some idea how long you might expect to live.
If you can shift to ‘half time’ or that age old standy, ‘consulting’ and still pull in some money, that’s good. Both for quality of life and emotional well-being and fiscal prudence.
If you are retiring still in good health and can’t live on 3-4% of your portfolio annually, I think it’s a potentially bad outcome for you. If you are forced to retire and have less than this, you can likely make it work…but not ideal. So…need to both 1. Drive down living costs and fixed expenses and 2. Increase risk and inflation-adjusted, after tax returns and portfolio size.
@Tom. From EU. Choice made by my wife on simple requirements: no earthquakes, no typhoons, no malaria, no rain season, easy resident card, no tax on foreign incomes, sun 80% of the time. Second choice: Uruguay. Probably better with children.
Edit: Malta is also a very nice place. And people speak English.
I have gotten a lot of great advice from people I really respect and admire. Sometimes I think I post too much but I enjoy it and, more importantly, I think I learn a lot when I post. This post is no exception regarding learning.
I was really struck with how it is very likely that a port will under-perform the sim. Not just for me but for the R2G ports. I understand that there are things we should do to minimize this but is seem like it still happens no matter what (on average). It is like a law of statistics. Then I began to wonder. Is it a law of statistics?
There is a law that behaves much like this: Regression (reversion) toward the mean. The sneaky thing about regression toward the mean is that it happens without an explanation. As Daniel Kahneman points out in his book “Thinking, Fast and Slow” no explanation is needed for the true observation that highly intelligent women marry men with lower IQs on average. You can come up with all the explanations you like but it is just regression to the mean.
Of course, there is quite a difference between a real company’s stock price regressing toward the mean and the predictive value of a set of factors in a ranking system regressing toward the mean. Or is there?
3 months (or a year) is too short to make judgments. As people have said already.
I think you need to break the question into (at least) a few parts:’
IN SAMPLE
Does the underlying factor appear ‘real’ in the sample period? Is it well defined? And over how long a period has it been stable and real as a factor?And how large and stable is the return for that factor. The expected ‘true alpha’ of a ‘rough version’ of the factor. Just for arguments sake, we can use a 50 or 100 stock ‘screener’ with a ‘relatively’ simple ranking (paired down factors) with no weight optimizations or market timing or hedging to test this. This is the ‘bench’ for the factor in sample.
A decision on how much ‘extra’ risk are we willing to take to try and capture the ‘uppermost’ tail of that factor with smaller port’s, higher-turnover, lower liquidity, more rules, optimized weights on rankings, market timing, hedging, etc.
OUT OF SAMPLE
3. Is the factor still appearing as ‘real’ out-of-sample? This can be tested with the same ‘benchmark’ screener that is, say, 50 or 100 stocks, with no optimization on the ranking and no market timing for a given ‘style’ of system.
4. How much is the ‘drift’ of our live money, smaller, more optimized port adding to or hurting the performance from our ‘baseline’ system in 1 (the ‘true’ style index for this port) out of sample?
Over time…in doing this, we learn…and we may start to see that some of any system’s over/under performance is because ‘factors’ rotate, surge-fade…have cycles, etc.
Some is ‘random’ market noise…and news / events.
Some is because optimization and increased rule numbers inflates performance expectations in ways it’s hard to initially understand.
Some is from the inherent costs of running the system itself (like Wuu Yean points out)…and
Some is because of the variance of extreme versions of the system around the ‘true factor’ performance. Both on the upside and the downside.
That’s my rambling on this. Factors can fade. But…some have been remarkably persistent. I think the amount of alpha in a system is probably inversely correlated with the speed at which the factors are likely to fade…because the systems tend to need to be more complex to get that high level of alpha…so the embedded risks get exponentially harder to identify and understand.
I agree…if…by regression to the mean you mean the ‘mean’ of all the possible combinations and weights of those factors within the system.
If I trade 2 value factors. And I tested 1,000 weighting perm’s to get the weights in the final system. I should expect that, regardless of what my ‘best’ version shows in-sample, the out of sample results will be ‘much’ closer to the mean of all the possible combo’s I tried. At…least that’s what I think.
