If you have not yet read this post, please do so first, and try to draw your own conclusions about which charts are random and which are not.
So the lesson here is that random data can look a lot like market data, and a lot of the patterns we (think) we see in markets can actually be generated by randomness. You can interpret this in many ways, but for me the lessons are that markets are a lot more random than we think. As traders, we really do not have an edge very often. (There is no edge possible in a purely random market, contrary to some of the suggestions in the comments to these last posts. For instance, all a good money management strategy will do in a random market is cause you to lose money more slowly as the vig (slippage, commission, the spread, financing costs, etc) slowly chips away at your capital.) It is, therefore, extremely important that we wait and pick our spots well.
As you struggled to sort out which charts were real (and if you haven’t done so, please do it before skipping to the answers), I hope you made notes for why you thought some were real and some were random. In my opinion, this is where the lesson is. There are, in fact, slight non-random elements to markets that make this exercise just barely do-able. For instance, volatility is often identifiable as a non-random element. Volatility shocks (an academic term, but a useful one. Think of a quiet market reacting to sudden news.) tend to persist. What this means from a practical standpoint is that the ranges of bars usually fluctuate in less-than-random ways: A sudden large bar will usually be followed by more large bars and vice versa. (For those of you who like math, there are a number of price models that capture this tendency in various ways.) It is extremely odd to see a very large bar followed by a tiny bar… but even then, anything can happen. If you haven’t done the game yet, approach it with that tool in your toolkit and see if it helps. Two approaches I have seen people try that will probably mislead you are to look for runs of white or black candles, or support and resistance. These appear just about the same to the naked eye on random charts as real charts.
Just so you know, I generated these charts as part of another project, but I had someone rename the files so I would not know which were real and which were fake. (Why? Because I thought it would be fun to try to tell the difference. We can debate my definition of “fun” later…) Here are my notes from my attempt to tell the difference. I correctly identified 4 of the fakes, and missed one. The fakes are: A,B, F, G, and J. Here are my guesses and my reasons. I am not saying this is the only way or the right way to think about it, but maybe my thought process will be interesting to some of you:
A – I guessed FAKE and was correct. There is a bar where H=L=C=O, which should never happen in an active trading day. This is theoretically possible on a very low volume (maybe preholiday) session, but extremely unlikely. At any rate, the rest of the chart was too active for this to have occurred. This one was a free gift. Had that bar not existed, I still would have guessed fake due to the presence of tiny bars near the extremes of some large bars. This is a price pattern that just does not happen very often.
B – I guessed FAKE because of the structure of the long string of white candles. This suggests a trend that picks up steam and then slows down… very unlikely to have this series without a black bar in real market data. Someone else who did this exercise said that this chart had to be real because you would not have 12+ bars close in the same direction due to randomness. This, actually, is one of my main points in this whole series of articles–it is very easy to underestimate how “streaky” random data can be.
C – I guessed REAL (without a ton of confidence) and was correct. I think I see typical consolidations and breaks through the low of the day, and the multiple strings of black candles with long consolidations in between also feel “real” to me.
D – I guessed REAL (with high confidence) and was correct. This is extremely typical trend day structure, inflection points come at the right time of day, and it just feels real. I know this could be random but I’m willing to take a big bet that it’s not.
E – I guessed REAL and was correct. Not a lot to go on here, but the 10:00 bar looks like a reaction to an economic number to me, so that’s enough to go on. If I didn’t have the first part of the day, I don’t think I would have gotten this.
F – I didn’t vote on this one, couldn’t decide. It was FAKE. On one hand, I see that the volatility profile of the day is bizarre. Very quiet morning, and then it goes crazy, but the afternoon looks like a reasonable reaction to that. I don’t know, maybe this is reaction to some very strange news item. I also think I see some short-term patterns that feel real to me. If I had to vote I would have guessed REAL and been wrong, but I just didn’t make a decision on this one.
G – Another O=H=L=C candle, so another giveaway. This one is obviously FAKE. Ugh… this is a problem I should correct if I ever do this again. Regardless, it is very unusual to see so many “violent” up bars followed by down bars, etc. I would like to think I would have seen that anyway, but as a valid test, this one is shot because of the little doji.
H – I guessed real and was correct. Classic trend day in every respect… so many little details that are unlikely to be random. Could be, but not likely.
I -I guessed REAL (with low confidence) and was correct. I am thrown off by the fact that the inflection points and shocks seem to be at random times of the day, but I also see such clear trend structure that it feels real. Most importantly, the volatility profile looks right… tightening consolidation, followed by breakout which leads to trend.
J – I guessed FAKE and was correct. My notes say “too weird”, and that pretty much sums it up.
So, there you go. Hope this was fun (and sorry the answer key was delayed!) I think another lesson is that it is much easier to tell trending real markets from random data. Randomly generated data will have trends, but there are some slight differences that will often be easier to catch than in trading ranges. There is a reason that I keep saying the best technical edges are to be found in trending markets rather than ranges….
12 Comments on “Randomness: The Answer Key”
Thanks, a very interesting question.
Actually, at first I was suspicious about F too. F will looks more real if the morning huge bars are placed around 2 PM. However, I finally decided to vote H. Being habitual contrarian and second guessing are the problems that I am trying hard to eliminate.
Jack
I really enjoyed this exercise. Thank you.
I also correctly identified 4 of the 5 fake charts. Considering that by any means I am not exceptionally good trader at this point, the fact that I still recognized 80% of the charts is another prove to me that market movement are not random and “efficient” and a trader can have an edge.
Interesting exercise.
If you had to select a single “point” for traders from this blog series, would it be that “your edge as a trader only exists when you are basing your decisions on information that cannot be random”?
This was interesting Adam. Looking forward to more from you.
We as traders take so much information for granted and I think your unique perspective to dig deeper into the meaning of price action is crucial to developing the trader mind. Thank you.
Yes, or a slightly different slant on what you said. We must be certain we are not basing our decisions on random patterns. The same point, but a slightly different emphasis.
If we base our decisions on random inputs, our results will be negative since there are costs associated with each trade.
I’ll post soon on risk / reward ratios and relationship the probabilities.
Goal 1: Recognize what markets to trade in. Based on the test, pretty much accomplished.
Goal 2: Recognize what setups to trade. Pretty much pending!
This blog series has been fantastic. It really provides an instant benchmark to evaluate almost any trading strategy. Simply asking the question, “Would this method trade a random and real barchart the same?” is a quick way to evaluate the vast majority of technical methods. Sadly, most methods *would* trade random and real barcharts the same! Outstanding Adam.
Yep. You actually just summed up the point of a whole lot of statistics and higher math in one sentence.
Thanks!
“There is no edge possible in a purely random market […] all a good money management strategy will do in a random market is cause you to lose money more slowly as the vig […] slowly chips away at your capital.”
Is it possible to beat a random market if there wasn’t the vig? I am not an expert in math or trading, but I thought up this simplified example: what if every day the market started at 0 and there was a 50/50 chance it would move in a continuous manner to either $1 or -$1. With risk management, couldn’t you then profit $1 during a random up day and cut your losses to be less during a random down day?
If the problem with beating a random market is in fact vig then, aren’t we then assuming that in ALL cases vig > long-term risk/reward? Is that a safe assumption?
The problem is NOT the vig. The vig makes it a negative expectation game, otherwise it would be breakeven.
To use your simplified example, how could you “cut your loss to be less during a random down day” as you suggest? Imagine that you are back to the coin game… how do you make $1 on good flips and manage your risk to lose less than $1 when it comes up tails?
In my opinion, if this idea that there is no edge possible in random markets is not clear to you, you will have problems thinking about trading issues with a probabilistic mindset. The same applies to using random entries in non-random markets… your results will be breakeven minus the vig.
Maybe I’m missing something in your question and assumptions, but I think this is a pretty important point.
The main point of my question is whether using a coin flip is an appropriate example for the markets.
When using a coin flip as the example, the coin immediately takes the value of $1 or -$1.
But do markets immediately reach a random value? What if we assume that the market will spend the whole day gradually trending up to $1 or spend the whole day gradually trending down to -$1? Whether the market goes to $1 or -$1 for the day can be random and a 50/50 chance, but you could still take advantage of the fact that markets don’t necessarily immediately reach that price.
Did that clarify my question at all?