Do Moving Averages Actually Work as Support and Resistance?

[@PaulSmithGsy]

Really nice to find someone who shares my passion for testing common assumptions that most people have no data to back up! Nice work. Thank you for the shares.

[@soreqt4736]

I like your works, man keep going, it's really great !

[@snay6869]

you are a real quant man

[@pranavswaroop4291]

Really interesting lesson for a noob like me, and exciting outcome! Thank you so much, wish you much good Karma.

[@leomelo2478]

Hi, man! Very nice work. Thanks for sharing it. I have a question about your method in this video. Do you count every entrance on the bands as an event or only the ones where the price crosses the average? If you count every entrance, there is a natural tendency for bouncing, since the price is already closer to the side it entered from. I think that, if you only count the events in which the average is touched, you have a better way to check if this effect is real, since a statistical test for the bouncing probability being higher than 50% would be simpler and effective.

[@neurotrader888]

Every entrance in the bands is considered. For example if the average was at 100 and the price dipped to 100.1 then bounced that would count. Most would consider this scenario a 'support' bounce. You are correct about the price being closer and naturally causing a bounce percentage above 50%. The same bounce counting method was used on the permuted series for the MCPT. If you look at the histogram shown at 5:44 the center of the distribution (the bounce percentage of the permuted series) is slightly above 50%. However in the real series the bounce percentage is far beyond what was seen on permutations.

[@boring91]

Great video, just a quick note, some of the plots are actually ETHUSDT not BTCUSDT

[@Gingeey23]

Interesting video and outcome, appreciate the reference to the scientific literature!

[@boring91]

I have a question regarding generating random permutations, I can see that the close price is generated by taking price changes and getting the cumulative sum. How about the high and low prices? We need them to compute the atr and we cannot use the same technique as we did the close price since they will generate inconsistent numbers.s

[@neurotrader888]

To permute bars I use an algorithm described in the book "Permutation and Randomization Tests for Trading System Development" The high, low, close are found relative to the open for each bar. Using log prices, it's close - open, high - open, low -open. The difference between the current open and prior close is also found. These relative quantities are shuffled. The first open price in the original dataset initializes the first permuted bar. Getting the high,low,close by adding back the relative quantities. The next permuted bar's open is set by using the difference between the open and prior close value.

[@snay6869]

THIS IS AMAZING BROOO

[@victorbozhkov7121]

You have to test if a upward (avgc10.1>?avgc10.2) trendline holds as support and vice versa for down trending. A sideways moving average or congestion of moving averages e.g. avgc10 != avgc20*1.05 is going to give you chop and the average does not matter since you are not in a trend

[@FreeMarketSwine]

I would expect any support/resistance strategy that deals with bands to have a >50% bounce rate with the bounce rate being higher with wider bands. The wider the bands, the further from the other side the initial entry into the bands is likely to be. Without knowing exactly how you tested it, I'm surprised that the random permutations didn't show the same results.

[@wong4359]

Your video is so good!

[@MA23312]

Hi, can you share the code would be appreciated

[@ventiladorbueno1846]

Thanks

[@renemiche735]

Gold , thanks a lot .

[@JaymeMGonzalez]

Awesome videos man! I have the impression that moving averages are more respected when the price is trending than when in a range. How could that be tested? Is there an approach you would recommend? I'm a beginner 😅 please be gentle if you reply

[@neurotrader888]

You need some way to quantify when the price is in a range or not. There are many ways to do this, all with various pros/cons. One way is to look at the autocorrelation of price changes. The autocorrelation tends to be lower during a range and higher during a trend. Not a perfect solution, but it's something. The video I'm doing next will be using meta-labeling (adding a machine learning layer to predict the outcome of a trade) to improve a moving average as support strategy. The same idea in this video. If you can wait a few days that video will probably answer your question a little better.

[@JaymeMGonzalez]

@neurotrader Awesome! Yeah ofc I can wait, I tried in the past using indicators like atr and adx but the results weren't so good. I'll look up autocorrelation in the meanwhile. Thanks for the great content 👌🏼

[@sarkash2978]

Every thing is Liquidity.. Ema are for retail users

[@jonathongeneral3744]

Your a retail trader

[@Neuroszima]

bro all i saw is just plain support and resistance taking place within first couple charts. The SMA applied is so vague, you can pick any "lookback number" and just watch for any "bounce" that happens to feed your own confirmation bias. This is worthless indicator. Also, why do you transform price into Log? Seems counterintuitive, and your hypotesis should include the thesis that actually traders use this exact representation to judge their decisions. Function that showed SMA, represented exponentially, when mapped back to standard coordinate system, would look differently than on a standard chart that we get to see when interacting in actual trading systems. (well lets put it like this, it will no longer be simple moving average, unless you only use graphical representation of Log price, and not actual Log price to deduce SMA because this is then purely wrong asssumption, you probably should use somthing that would be "logarhitmic moving average"?? and not deceive potential viewer??)