I raised a point (not a criticism) that fundamentals don’t change much on an intraday basis, barring a few relatively infrequent events. Was that the criticism you meant, rawraw, or was it something else that I missed while skimming here?
Edit: maybe it was KMd’s idea that she couldn’t research enough on fundamentals to outdo the market? I can see what you mean here. If fundamental research can’t outperform, how could price and volume research (which is often easier to do) outperform, then.
My apologies… I did not intend to discredit fundamentals (especially on a CFA forum ) Both technically and fundamentally driven participants are crucial to making the market work. Fundamental analysis is very important. These participants are the ones that bring the facts to the table. Remember my initial example in my opening post. The market is a tray with sand and iron filings? Yea, the fundamentalist form the edge of that tray. REALLY IMPORTANT. However, technical participants are really important to. They digest the meal that the fundamentalist feed the market.
Technical analysis is misunderstood. All of it is different ways to infer the sentiment and behavior of participants. So, through technical analysis, I can indirectly make judgment about fundamentals since what I am looking at is how participants have reacted to them.
Time frame has a lot to do with why technical analysis is the favorite of traders. Fundamental analysis is useless on an intraday timeframe. On the other hand, I would not recommend technical analysis for long term investment choices.
My final comment, about the market knowing more than I can research, is a personal bias. You can do your research perfectly and have everything in reality play out how you expected. However, you never know how the world will interpret that result. That is where TA is really neat. Sometimes it will show you, hey sentiment (because that is exactly what it measures) is seems stunted even though the price is up. Then, when your well researched good news comes out that is supposed to make the underlying 5% more valuable, you won’t be surprised when the price actually goes down. That is a horribly oversimplified example, but my point is how participants are interpreting fundamentals is something only the market can tell you.In other words… don’t discount the validity of collective consciousness.
Sorry, I went on and on there
About algo vs discretionary. The difference is that algo is back testable but discretionary has a little “art” mixed with the science. My rules might be to see 3 reasons to take a trade (or something like that). Those 3 reasons could come from a basket of 6 things I look out for. Even then, I can use discretion in whether I want to pull the trigger. Maybe news is about to come out or there is something going on in a larger timeframe that looks like a warning. With algo… you take your signal, no questions asked, because your edge is based on you ALWAYS taking the trade.
Both are rules based. Neither is just hacking away at the market on gut feel. Discretionary is just a little more loose.
Okay, then that is good we agree. So my question is given the well documented short comings of the human mind in making decisions, why would you think the looser approach is better? I just can’t understand how a rules-ish based system being executed via the human operating system would outperform a rules based system. Only thing I can think of is perhaps there is a large amount of experience and knowledge below consciousness that ultimately contributes more than the many shortcomings of our advanced monkey mind. But since you are new to this, surely this isn’t the case?
I love this conversation! So, I would not say that one school of approach is better than the other. The both have their advantages and disadvantages. I actually started out trying to be an algo trader. I was working with a programmer and we had some really great trading systems that we worked on together. The scary part is when you start back testing them. We had this one system ( which we affectionately named “penicillin” given that part of it’s charm came from a coding error) that… no exaggeration… made a few hundred percent (like 500%!!)return on capital/ year in back testing of recent years. The problem was, before 2013 the thing blew up. It would lose a couple hundred percent of capital in some years! The problem is, you never know when that is going to happen again. The market is always changing and even though there are ways to forward optimize, given my time and resources it just became impractical. Algos can be very powerful but I’ll leave it to the quants with the time, resources and capital to make them work. However, at the end of the day,… y_ou can only make these algos so “smart” because they have to be based on if/then steps_.
Below is an actual discretionary trade I took last night in gold (I cherry picked a winner, sue me ). There is no way you can build an algo to take the trade that I took. Each trade I take is so unique, even though process based.
It all starts with the context of a volume profile, but not just any VP. It was a composite I specifically chose because it covers all the price action from the last low to the last high. What kind of composite I use is itself a discretionary decision based on what information I am interested in getting out of it. Already, you could not have an algo do this because “it depends”… it is not “if/then”. Moving on, although I have a game plan I am looking at, I am also continuously “reading” the market which adds to that thing I call context. Really, you nailed it Rawraw. The advantage of the discretionary trader is that in relying on skill and practice, they get to trade with their eyes open. The algo trader is essentially blindfolded and at the mercy of an ever changing market.
While I can see the human element adding some edge over straight algos, it also certainly adds the ability for additional blow ups and errors than the algo. The only intraday trades I have ever ventured to make usually just involve ridiculous market moves on “news” that is really nothing when a stock drops 10% by 10am I usually pick up a bit and sell around noon when it picks up a bit. Can pocket a quick 2-5% majority of the time (however have been pummeled on those trades as well)
Still the intraday workings of the market are too random for my likes. I find the workings of the market & market micro structure very interesting but I am not dipping my toes in that pool, the learning curve seems way too steep and the participants you go up against way more skilled than some guy who worked in IB at GS and now is a PM. I like my odds against him over the many PhDs that are way more well versed in mathematics and designing systems to pick off penny’s than I could. I also suppose thats why your discretionary system has a possible positive return, the human element ads something their algos likely could not do. If you were just designing an algo system it is unlikely you could surpass others out there who co-locate and have much more in terms of resources.
Anywho, that is it for my ramble. Will be interesting to see when you go live and would like to continue hearing updates.
Thanks. I do expect to be successful at this. It will just take time. However, just in case it does not pan out, one thing I learned from AF… I can always just become a plumber
I am still not convinced of your premise that algos are bad because you are strapped to a nuclear bomb. It seems you are focusing on the investing algo part. But what about the position sizing and risk mitigating part of the algos? Algos aren’t just blind to risks, they do what you tell them to do. Are you ignoring the latter or was that included in your rules?
I think investing (whatever the sort) can be simplified into probabilities. Different situations produce different probability of outcomes. In fundamental analysis, spinoffs increase the likelihood of finding a mispriced business due to market structure. This doesn’t mean every spinoff is a mispriced value investment, just the probability of finding one is higher than just looking in the S&P 500 group.
I have limited exposure to trading and it all seems to come largely from those Turtle guys, since people always talk about them (http://bigpicture.typepad.com/comments/files/turtlerules.pdf). But it seems to me it’s the same concept. Certain things increase the probability of profits, so you spread lots of bets over those trades knowing with a large sample size you come out ahead. And this is why the huge flaws of our minds are important. There is no guarantee that you can repeat that trade above, or even be aware the reasons why you made that trade. I think it is going to introduce a lot of noise, which again a fault of our brain, that you may take as progress or failure and really it is neither. The key here, just so i can be explicit in my point, is the frequency, speed, and number of decisions required in trading vs fundamental analysis. It really amplifies the problems, all else equal. So while the brain is flawed in both, my point is trading is much more prone to experience the glitches.
Without knowing anything about your blowup algo, I’d assume it was either concentrated bets or didn’t focus on downside risk management. If that’s true, I think it means the algo is incomplete and not that the algo strategy is lacking.
My experience in this kind of stuff is that position sizing and stops are actually the more important part of trading system success. Van Tharp did an intesestig demo that one could have a system that was near random or random in terms of technical signals, but could slowly build up equity if the risk part of the system is done right. A lot of traders talk a lot about the technical signals without talking about how position sizing and stops work, but that’s the part that ultimately makes or breaks the system, particularly in terms of blow ups (vs slow declines).
I will start by again saying, I don’t necessarily think discretionary is better that algo. I just think discretionary is better for me given my resources and capital.
I think to be clear, I should spell out how an algo is developed. For this discussion I will use a very lame trading concept for the sake of simplicity. Lets say you have an idea to trade off of moving average cross overs. You have a “fast” moving average (MA) (based on a smaller amount of periods) and a “slow” MA (based on a larger amount of periods) When the fast dips below the slow, you sell. When the fast crosses to the top of the slow, you buy. Here is where you go from there.
You prepare code that that can run on historic data and essentially trade by the instructions you give it. Embedded in this code, you have a set up where you can tweak the parameters of the instructions. For example, you might want to test a 30pd MA for fast vs, a 50pd MA for slow… or maybe you want to try a 30pd MA for fast and a 55 for slow… and so on. You can see how complex back testing an optimization can get if you have too many degrees of freedom.
risk management… even more parameters. In addition to tweaking the settings on the trading signals themselves, you will also test different ques for profit targets and stop strategy. Do you want to go for lots of small losses and a few big wins (positive skew)? Do you want to aim for lots of small wins and a few large losses (negative skew). Do you want the stops and profits to be based on signals or percents, or ticks… or even something more exotic like some multiple of average true range?
The actual testing. The proper way to do this is with something called walk forward testing. This is the best solution to prepare to have the best out of sample performance. First, you need to pick your test window size. For an intraday system, like I was testing, a one year test block is fine because it contained enough trades to make an adequate sample. Then you go back to the earliest time you want to start testing. For example, you could go to 2010 and start testing. Now is when it gets complicated. You ideally will run all permutations of parameters and pick which works best. Not only which works best, but which set of parameters is most stable. You want to find a set of parameters that not only performs well, but the neighboring sets also do ok. This helps draw the line between over fitting and optimization.
Once you have the best parameters for 2010, you will “walk forward” those parameters onto 2011 and see how they perform. Then, you run the parameter optimization drill all over again for 2011 and “walk forward” those results onto 2012… and so on. After you walk forward all the way to the present time, you will have good idea of how the algo will perform going forward and what kind of draw down you may be dealing with. That is critical to know because that is how you determine how much capitalization running the algo will require.
what do you look for in optimization? As Bchad said, net P/L is a weak piece of data to base the decision on. You are going to look more for things like the comparison of max draw down compared to max draw up. You are looking to optimize the sharp ratio, not necessarily the bottom line.
I have tried to spell out all I know on this subject in a few paragraphs so not a complete explanation. However, you can see how complex algo trading can be and if you try to make them more “smart” your degrees of freedom will skyrocket! I’m sure at the instituional level they have elite staff and computing power to handle more complex algos than the one I used in my example. For myself, what I can do as a discretionary trader is much more powerful than what I could get out of an algo.
Let me take two steps back. I think we are using two different definitions of the word algo . I’m talking of any rules based system. Of course this can be executed programmatically, but I view that as a subset. Admittedly it likely reduces the temptation to deviate than manual execution of rules. Do you have a list of rules that you are manually implementing? If I was on a mission to become a trader, I’d want an explicit list of rules that functions as my hypothesis. Then I’d want to track how they work. It seems like you may be Journaling, which is a qualitative form of this .
BTW I don’t have strong feelings about this given my lack of experience . Just helping your discussion out given the lack of traders around here ha ha
I’m happy to go on and on about this…and I will . I don’t think I have made clear the difference between the algo approach (computer automated or manual) vs. the discretionary approach. They are actually VERY different even though both process based.
They are different because they are getting their edge from two separate foundations.
In algo trading, your edge is that you have found a set of events that have been probabilistically proven to have a favorable outcome. There is no room in that sort of trading to meddle or deviate since your edge comes from that exact set of rules. Therefore, it essentially has to be something you could program and automate since in order to have an edge, it must be back testable.
Discretionary trading gets it’s edge from the trader being adept at reading the market and having developed risk management. This kind of trading is process based, but not an explicit set of rules like in algo. Discretionary trading is more like a detective building a case. Yes, the detective has a loose process for how he gets the job done, but it is not an explicit sets of rules. It is all based on the individual case …the context. You could not, for example, say to be a detective…1) meet with a crying woman 2) find a bloody knife in the bushes 3) take finger prints of the woman …and so on. What you can do is say… 1) describe case 2) interview relevant individuals 3) analyze evidence 4) do research… and so on. Similarly, for the discretionary trade I posted above, I don’t look for those specific items. I do this: 1) define context 2) collect evidence of market participant’s behavior 3) identify opportunities to risk small for a larger profit 4) manage position based on price behavior.
So in closing… Algo trading gets it’s edge from probability and discretionary trading gets it’s edge from skill.
I have a tactical asset allocation model that is algorithmic. My attempts to layer discretionary trades on top of it have not been all that valuable, and I concluded that it was best to follow the system. Then again, until this year, it’s been hard to outperform 100% long stocks since 2009.
I think the challenge is that we overestimate our abilities to see the context differently and more valuably than others.
If I were to go back to doing discretionary stuff, I think I would use algorithms to manage my risk and position sizing, and put specific time frames on my discretionary positions such that they would either be validated or confirmed wrong by a certain date.
It’s basically a trend following system that uses moving averages and a flavor of risk parity weighting with a few extra bits thrown in. We try to be careful not to over-optimize the moving average windows, though others who have looked at the system can’t resist trying.
It has not been performing well lately (mostly because stocks and real estate are the only things that have really been doing anything since 2009, and this approach dilutes those returns with other asset classes). I’m surprised that my business partner still wants to run it, but it is good at saving one’s butt in a bubble/crash situation and generally drags by a lot when late in the market cycle. It can be levered a bit if the client wants to, and it is a decent diversifier for other more risky kinds of strategies.
My main concern about it is that I think more value comes from the risk weighting part than the trend following part, and the forward prospect on risk weighting looks less sanguine because of fixed income being at such historically low yields. Like everyone else, the concern is if interest rates trend upwards (as they eventually must), it could make every asset class go down together.
Plus, risk weighting tends to magnify the effect of fixed income on the portfolio, which is why it backtests so well in the 1980-2010 period, but looks riskier going forward. On the other hand, if everything does go down together when interest rates creep up, the system will probably end up holding a lot of cash, which isn’t such a bad strategy for that particular state of the world.
It’s been a while since I actually did any improvements on the system. This is partly because it is used in production and other people want the rules to be stable. But I’ve learned a lot since we first designed it and perhaps it’s worth revisiting whether I would recommend designing it the same way if I were to start it from scratch today.
First of all, I would like to go on the record as saying… there is no crystal ball! All TA can do is help you derive evidence to answer questions like “What is the market trying to do?” “How good of a job of that is it doing?” “What might it do next?”
So here is Mexico…
The volume composite is of all of the trading volume since 2014. Of course, since this is an ETF, the fact that there has been some minimal decay should be taken into consideration (although… most of the volume for further back is higher up so does not effect this lower part of the profile anyway). So, what is the market trying to do? It is ranging and trying to price in these outcomes that could be very material for Mexico. I labeled the early year crashes because that was like a reset button for markets. Price discovery had to start over again after that. As you can see, the prices being considered for the current set of fundamentals are found in range #1. However, a surprise would see possibly a push over that “low volume node” and into area #2. However, most likely that LVN will act as resistance because I can’t imagine Hillary would be that awesome for Mexico’s fundamentals. The yellow line is the most highly traded price of the past three years. It is just context… maybe useful, not useful, or at least amusing.
) Both technically and fundamentally driven participants are crucial to making the market work. Fundamental analysis is very important. These participants are the ones that bring the facts to the table. Remember my initial example in my opening post. The market is a tray with sand and iron filings? Yea, the fundamentalist form the edge of that tray. REALLY IMPORTANT. However, technical participants are really important to. They digest the meal that the fundamentalist feed the market.
). There is no way you can build an algo to take the trade that I took. Each trade I take is so unique, even though process based.
