Why you should add to winners and never add to losers

July 17th, 2015, 01:48 PM

artemiso

This is not analogous to scenarios A, B or C. Do you see why?

In Scenarios A, B or C, you are making an action after the realization (market data) at time T+1. I gave the correct analogy to this: I said that you can precompute ahead of time that if the sun doesn't rise at T+1 that it yields an expectation of +$1M to play the game. So when the sun does indeed not rise at time T+1, then you would play the game. Your action time is T+1 and your decision space is {bet on rowing a 6, don't bet on rowing a 6}.

In the two opportunities you are giving me, you are only allowing me to make an action at time T (before the realization at time T+1) predicated on me already fixed on playing the game. In this case, the expectation values that you should be comparing are P( sun rises AND row a six | play the game)*6M and P( sun doesn't rise AND row a six | play the game)*6M. Your action time is T and your decision space is {bet on sun rising, bet on sun not rising}.

Apologies bad example on my behalf. Since the resulting game is the same in both scenario's I agree the "decision space is {bet on sun rising, bet on sun not rising}".

I should have picked opportunities with different resulting games to make it clearer. If I make it
Opportunity One. "If the sun does rise tomorrow, I will let you play this game where I will pay you $100 if you flip a head on a fair coin and $0 otherwise.
Opportunity Two. "If the sun doesn't rise tomorrow, I will let you play this game where I will pay you $6M if you roll a six on a fair dice and $0 otherwise.

I would argue that the expectancy of Opportunity One -> $50 and that the expectancy of Opportunity Two -> $0

October 11th, 2015, 04:24 AM

paps

Great post @SMCJB totally agree.

The opposite is only possible when only cash rich and a different mindset. I remember few points while a few years ago reading West of Wall St...the process was detailed how few pros avg down...but it calls for strict discipline and ability to take a loss if not working. Seen and done that...hence totally agree to your nugget.

Thnx
S

I too am a reversion trader who averages in as the position moves against me. It does require at times deep pockets, a sound plan, strict discipline and a very different mindset. You have to really believe in what you are doing.
Cheers John

October 13th, 2015, 08:56 AM

i960

One of the other issues in this debate is that it seems the extremes of 0% and 100% are being used to prove or disprove a point. As I said earlier, p is not necessarily static throughout the decision process. Perhaps what I meant by that was that they're independent probabilities occurring at disparate moments in time. They appear linked, but aren't.

If we want to get practical/pragmatic and look past the 0%/100% p stuff, when we consider things like S/R, ranging >80% of the time, yadda yadda, then technically adding to losers should actually outperform adding to winners over the long-term in the majority ranging case. Obviously in the hard core trending situation this won't be the case. I know this, when I add to a position that's gone against me, I'm not adding because I'm trying to dig out of a hole quicker, I'm adding because I still believe in it and still expect it to reverse. The initial position could be considered a "get in the trade" at a reasonable (but perhaps not perfect) entry point.

It also was never answered in the original post if adding to winners used the same stop as the original position.

I agree with you, and this has been proven correct in my trading for many years, as a matter of fact I deliberately do this in order to get the best RR in an uncertain environment. The only proviso is "Kids don't try this at home unless you fully understand when not to do it".

Cheers John

October 25th, 2017, 07:18 PM

I just found this thread and always found this subject interesting .I have been testing this for a couple of months. Both adding to Loosers and Winners and averaging them out against each other in order to close them out automatically when ever an avergage is in profit or break even. Sometimes it might close out 50 trades at a time. Over time with changing markets nothing of this will work. Just stay with fixed contract sizes. Over 5000 trades have proven it to me.

October 26th, 2017, 09:15 AM

The math here is above me but my practical take:

Favor not averaging down if:
Market does not mean revert or is not mean reverting (is jumping or trending)
Leveraged
Fixed or short holding times
Strong signal to noise (confidence)

Favor averaging if:
Not leveraged and you can place many trades
Market is mean reverting on multiple time frames
Capability to hold for high probability recovery
Low quality signal or high uncertainty (low confidence)

Another thing to consider is that there is uncertainty in trading when buying against the trend (for mean reversion). For example, I might think the market has a 20% probability at stopping at support 1 but if it does not then I think it has a 20% probability of stopping at support 2 but if it doesn't then it might have a 20% probability of stopping at supporting 3. This sort of analysis implies also that if it doesn't stop at support X then I expect it to continue until it hits support Y.

I have been using averaging in BTC for a type of naive market making. In this "game", you are never forced to take closed losses (you never lose money per say) but you will convert your dollars to asset your trading. You take your pile of money and divide it up into minimal lots of like 100 or however many you want. Now, you just place trades without stops. Anytime you think the market will do something you place a trade. It has to be a limit order. But the goal is to capture some factor of mean reversion or spread. It is sfaer and easier to buy then sell btc so the only downside is that your trades will be all longs. So some will be in the same direction. The purpose is not to average in but it has the same effect.

One way to imagine this is if you think the market will stay in fixed range between A and B. Assume it has a random probability of being between A and B and the FV is the mid point C. You do not know where it stop between A and B but you want to capture any deviation from C. So, you stack the book above and below C. Now imagine it makes fast cycles, you will replenish the book and capture the deviation many times. If it makes large cycles your average will be the average of A and C.

One of my original inclinations for such a strategy, was two things, first another type of market making program I read about. But, the other was that I noticed when I made predictions that all my predictions had value even when a new prediction conflicted. Aronson's Evidence Based Technical Analysis influenced my prediction ability early on: I quantified all my predictions. But, I had a nagging suspicion that his ideas were wrong or at least incomplete. They are. While it is reasonable if markets are efficient that the value of any signal will fall off quickly, this is normally the case, it doesn't mean it will go to zero. Also, if one uses stop losses then often the optimal stop loss required will be larger. So, in other words you have a type of asymptotically fall off measure for any prediction. As an aside there is a another thing wrong with the idea of falsifiable predictions, it assumes there is only one path. However, you can take a statistical view of infinite paths within set limits: this will yield your return. One of the primary reasons traders lose is because they have some inclination that is on average correct but they do not trade it in such a way that they can profit from the average outcome. For example, if I say the market is bullish near term. It might mean that the market will go up more then down. It might also mean that it will go straight up. Anyway, there are various paths. If you try to bound that with a stop loss you are likely to lose. However, if you divide up your capital into infinitely small portions and take every possible bullish trade possible and average them then you will know the true value of the statement. As an example, if I think the market is bullish. Imagine if I just go out and buy a futures contract with a stop loss. My stop might get hit and the market might then rally from my stop out. However, imagine I take my capital and divide it into X allocations. So, instead I might go out and buy a futures contract but also buy a bull spread in the options market and put in a limit order far off market. This allows me to profit from multiple paths. This is why also that granularity of trading is highly advantageous.

One style of trading I developed to take advantage of the insight that all my predictions were valuable was to simply look for opposite side trades while keeping a memory of my original idea. So, in other words if I predict the market will decline off the open but it starts to move against me. I will open short scalps against my position. This could be done in a correlated asset or by simply going flat. Anyway, when I'm done scalping against my position -- any mean reversion will allow the original position to recover. However, if you managed to scalp well then your closed profits will probably keep you close to net flat. I have read about gamma scalping in options but do not understand if the concept is similar.

This is a very powerful technique but difficult to really if highly leveraged like in futures. You need a larger account or a reduced leverage to make it work like it can. By trading against your position, you allow it to remain open for the maximum or optimal threshold while reducing your risk.

The naive market making theorem in BTC is similar. You have to pick the areas where the market is likely to mean revert. You can play a large number of "games". But, you never incur a realized loss because you convert your losing trades into BTC. That's one way to see it. Of course, if too many trades become "hung" or "lost" then your returns will simply mirror the returns of the asset.

There are a few keys with the market making game. You must be able to read the market or predict it to avoid getting hit by the trends. You also need to keep leverage very low and some capital in reserve. If the market starts to mean revert on any time scale, if you can scalp out profits in summation that exceed the hits then you will make money. It is a statistically sound method of trading. Of course, on the other hand if you lose then you do not lose money but just convert to the asset. On the other hand, most of my performance as been made by predicting high probability trends/jumps where I deploy 100%. For that style of trading, leverage would boost my return significantly and I would want to use a stop loss. This is why also there is an interplay. For example, if I'm market making then I will try to predict the other traders targets and that is where I will offer. So, there is a type of interplay: the market maker wants to keep size small in case the other trader has an edge or knows something. The directional trader wants to trade big to take advantage of the statistical probability of the trade. Also, it is somewhat mathematical. Most buys and sells in the market are let's say noise but statistically a small percentage will have value. So, if you're taking the other side you want to keep your risk small.

These are just my 2 cents. However, if you're market making you must also factor in the cost of the other traders market orders. Unfortunately, if the market orders are not the same cost to everyone then you subject yourself to greater adverse selection risk by the primary market maker(s).

Why you should add to winners and never add to losers

Discussion in Traders Hideout

Why you should add to winners and never add to losers