The Myth of Averaging

Averaging UP

Ok, new chapter in this exciting journey to prove the mathematicians wrong...
I have uploaded a new EA in post #1 that implements Averaging UP (adding to winning positions) instead of Averaging Down (adding to losing positions):
The strategy is as follows:

1. Make a buy or sell (chosen at random) at a fixed hour each day.
2. When price hits the ScaleInPoints level, another trade in the same direction is added to the position (keeping the previous trades open).
3. A trailing stop is used to close the entire basket of open trades at once.
4. Each level has its own trailing stop distance (used for all trades in the basket, not just the last trade)

All other features are the same like the averaging down EA:

- There is a RandomSeed input parameter to create different random buy/sell sequences. Setting the RandomSeed to 0 will open both a buy and sell trade each day. See post #1 for more details on this.
- The EA will only trade at the open of each M1 bar. All other ticks are ignored. This helps us to run thousands of backtests very fast.

Ok, let's start trying out this thing...

First let's choose a good time of day to enter the market. We optimize "Balance Max" for EntryHour:

All terrible performance and not much difference:


So let's choose EntryHour=10 for now and move on. We probably have to revisit this input parameter later...

Now we are going to hunt for a good L1_TrailingStopLossPoints:

This gives the following results:


Hmm, this is awkward. The best is L1_TrailingStopLossPoints=10, but we don't want to build a scalper EA that is trailing at 1 pip distance. Moreover, that setting only made 9 trades. Similar for 20 and 30.

Looking at 120 that makes more sense. It is at a local maximum in the graph and generates 2045 trades. Let's choose that one.

Next step is the Level 2 settings:

I am a bit disappointed in what this gives me:


Hardly an improvement over Level 1... The "best" setting is L2_ScaleInPoints=800, L2_TrailingStopLossPoints=375:

The algo in my sig adds into winners, I am testing it in real-time on VPS after successful back tests.

maybe check them out in 6 months.

Thanks. Good to hear you are able to add into winners. So far my experiments with averaging up are not hopeful as you can see.

Level 3 optimization gave: L3_ScaleInPoints=200, L3_TrailingStopLossPoints=375.


I am giving up...

Since AUDCAD is fairly ranging, perhaps this is not the best symbol to add to winners and follow trends?
Let's try EURUSD now...

Worked with different breakout strategies using EA in the past and found that USDJPY has by far the best trends. Maybe you should try it on this pair.

This is the best I can make of EURUSD:

At least equity is going North, but certainly not as convincing as the averaging down equity curves we have seen.

Thanks for the suggestion! Here is USDJPY with an Averaging Up strategy:

For reference, here is USDJPY with an averaging DOWN strategy:


Unfortunately both curves look correlated. Especially 2014-07 is a bad month for both with a strong peak in deposit load. Running both strategies together will not help us get a smoother curve.

What's next?

I don't know. I have now experimented with averaging down and averaging up. The results proof that even a strategy without an edge (a strategy that just blindly enters the market without any logic) can have a positive expectancy if we apply averaging down/up with the right settings!

Are the results good enough to trade live? No, I don't think so. As @ryuryu pointed out in post #16: this strategy needs a well-funded account, or very high leverage to make a relatively small profit. That is just not worth the risk. You can do better with other strategies or even a simple buy-and-hold of one of the indexes (e.g. S&P 500).

However, we did learn something from this: never blindly assume others are right, and always challenge mainstream ideas about how things work. Even mathematicians can be wrong sometimes. I could do a similar thread on grids or Martingale, and who knows, we might discover more unexpected things.

All in all, this opened my eyes that my usual strategies, where I stick to having only 1 trade open at a time, are too limited. It is ok to scale-in if the first entry was not well timed. It is ok to scale-in even more if the second trade was also wrong. As long as you choose the right scale-in levels that match the symbol you are trading and don't keep adding beyond some point, you have a good chance to close the basket in profit! These experiments will help me in tuning the scale-in and take profit levels of my other strategies.

There is no myth here. I've been following the topic from the beginning. I'll toss in my coin. I have been averaging positions for several years with success. I also know another dealer who has been doing this for over 10 years. Averaging works, but cannot be rigid. At some point, the market "runs away" and the farthest position must be paired or several closed. I agree that you need to have a large leverage and a lot of capital. Do you earn little? Monthly it comes out 15-20% so there is nothing to complain about. The averaging is carried out on at least several accounts. How to kill it doesn't matter. Act, think and something will come out of it

Last few days:

Such as limiting yourself to max. level 3 or 4 like my example? With "paired", do you mean opening an opposite trade in another correlated symbol?

Wow, that's impressive! The examples I have shown in this thread are no where near these numbers. However, I assume your entries do have an "edge" so the averaging only kicks in occassionally, not 50% of the times like in my examples?

Paired = this same direction but diffrent levels. If you are profitable close both (profit and loss) but on the bill you are in plus. Thats the point . Look at screen. Some TPs ale below buy- they will be closed with "loss". Why ? To be closed and not left for better times. If you don't close them you will blow up account.

In june I have +14,29% so not as good as I want because of low volatility. You have to find pairs that works in consolidation or often turns back as GBP, JPY, EUR[USD], Oil, Gas etc.

I see. I think what you call "paired" is what I call "basket", right? A set of trades in the same direction on the same symbol that are typically closed all together at the same time.

Yes, that's this same. It have some risk because of no SL so you have at least 2-2,5k $ on account per 0.01 but it works.

You did some amazing research. I appreciate you sharing that with us. I wonder what would happen if we'd combine it with some form of directional edge. Even if its a little. I guess the mathematicians aren't wrong, but aren't completely right either.

I've only skimmed this thread because I'm on my phone, but I don't think you're looking at averaging in the right context. This appears to be based on a random mathematical approach and I agree that it's unlikely to be profitable long term.

But used as a method to manage a losing trade in a profitable system is a different thing altogether. I can generally pick which direction a pair will move accurately, but my entries are often badly timed. By averaging I can make my losers become winners.

I've backtested this theory and it still makes the occasional mega loss, but over a period of 5 years it was more profitable than closing losing trades. I'm still not convinced, but I don't think a random mathematical model is necessarily the right way to test this.

What I tried to check in this thread is to see whether averaging actually does anything useful. Of course a profitable strategy with an edge would be better. However, given that even the fully random strategy I tested can be profitable, this seems to suggested averaging does indeed improve your odds. So averaging is clearly a useful component to include to your strategies. See also post #52.