Another Trading Journal

Hi,Unnamed Playr..thanks for your threads. I'm new here and i've been trying to spot out and understand some stuffs in the market to improve my trading experience. I'm kinda so impressed with what you are sharing..it's so insightful and helpful. I actually decided to start from your very first post way back 2010 so as to squeeze out every insight you've shared. Thank you UP.

HI!Where you find data?As far as I know they are updated every year.

HI UP!Why Constant factor "1" ? For an example: ECXt = 34.38805726 x (USDt) – 0.3155 x (GBPt) 0.3056 x (JPYt) – 0.1891 x (CHFt) 0.1113 x (SEKt) 0.0785 What is the formula to calc "constant factor"?

Hi UP,

I wonder if the imbalances you are observing are mere reflections of your broker's books. I attached an example showing NZD and JPY indices, the direct NZDJPY pricing as well as a 50:50 coefficient pricing via CAD. No matter what combination I key in, the resulting imbalances in comparison to the directly priced pair are minuscule for me - I am using a Pepperstone ECN razor account. Thanks for your answer, in case you are still around.

...or even the result of a calculation of non-synchronized bars. I noticed the USDIndex indicator you referred to having that problem.

Please advise.

...or funding/speculative pair EURUSD on M1.

Hi UP,

I think I figured something out - sort of.

So in my outsider / dummy words: I kept thinking how incredible a beast these market dynamics are, given that there is an equilibrium with the maximum uncertainty of the spot Bid-Ask spread most of the time - otherwise the HFTs (including myself) would arb the hell out of it. However, something must give in order to keep the market liquid and flowing that efficiently and that means the big players do take stock risk.

Stock and flow, that was the AHA-moment for me.

I then finally managed to overlay the percentage changes of underlying currency indices over the relevant pair and, while still in perfect correlation and similarity, I started to actually see the spreads opening. Those spreads drift and fluctuate and pose pretty much the same problem like any other financial time series, but: we don't need to catch the whole spread, we can just cut it up and "renko size it". Therefore, I coded a separate spread indicator and tweaked it to snap back to zero when x pips excursion (=arb opportunity) happened. As such I "forced" mean-reversion rather than waiting for the fluctuations to carry it back. What I initially thought to be a regression challenge in terms of sample size, etc. turned out to be me making things difficult: the lot sizes are the coefficients of the whole construct, e.g. 1.0 lot EU = 0.5 lot *EJ x -0.5 lot UJ - and it doesn't matter which pairs you chose for the hedge other than picking those pairs with the lowest combined transaction cost (B-A spread+commission).

So I started testing entries, re-entries and exits when the excursion was 3 times the combined transaction cost, e.g. 3 x (0.8 pips+1.5 pips+1.2 pips)=10 pips. I have not come across any situation I did not get out profitably; some are 2-3 pips only, others at >20 pips. The logic was to exit, when the spread started to go against me the first time after the snap-back / forced reset to zero. And the most awesome part is, that it can be automated entirely.

Maybe, just maybe, I finally found my personal exit door from the horrible rat race of traders' discretion and screen boredom. Off to writing some code. Thanks for revealing the bits and pieces that helped me form my own picture.

One correction:

It does matter which pairs to pick for hedging other than just their transaction cost: One must avoid pairs correlating too much with the main pair.

e.g. see picture

1) UJ reaches trigger spread against all USD und JPY pairs: main pair to be traded long at 1 lot.
2) Pick a USD pair with least excursion - e.g. UC, to be traded short at 0.5 lot.
3) Pick JPY pair with least excursion - e.g. AJ, to be traded short at 0.5 lot.

Back to testing.

Going through the tests and coding iterations, I noticed a few pitfalls which I mention here in case somebody is attempting something similar:

1) Synchronization of bars. The way I'm doing this is by specifying a Time from which I measure forward. I chose the previous weeks close at London time. Bar charts have holes and there have been cases, where the brokers change the history. As a consequence some charts have x bars whereas others have x+-y bars for the same period. However, this method allows me to achieve sufficient synchronization at the current bar and results can be reproduced.

2) Excursion limit or entry trigger. I realized that transaction costs and pip distances are inconsistent, if applied across 28 pairs. That's why I take a Dollar figure and convert it into pips for each pair, e.g. whereas AUDUSD needs 10 pips to trigger an entry for 100 Dollar, EURAUD needs 13.2 pips.

3) Every hedge starts with a primary pair. This pair is chosen based on reaching a) the excursion limit and b) being the most liquid amongst all the pairs breaking away (e.g. all AUD pairs could reach the limit at the same time). Since the excursion limit means the same dollar figure for each pair, liquidity is measured in excursion / avg transaction cost per pair.

4) The primary pair remains the focus until the closure of the hedge. The pairs picked to hedge the primary pair are selected based on stickiness to the mean - e.g. if AUDUSD is the primary pair and AUDJPY has moved the least of all AUD pairs, I'd pick that one. The same applies for the USD pairs.

5) In case the primary pair keeps drifting against trade direction and a new excursion limit has been reached, I re-enter with the primary pair, but for the hedges it might as well be the case, that different AUD and USD pairs will be chosen. As such, I might end up with a basket of 5 pairs after 2 entry sequences.

Cheers

Alright, I think I have some ideas on what unnamedplyr is doing, however I am not sure about something he said, and would love to hear everyone insights.

1) He mentioned that bank offset inventory exposure, and hinted they do it in the largest market, the largest market is fixed income market, let assume bank accumulate a large USD inventory and USD value is going to tank, the bank offset it in bond market? this doesn't make any sense as any fixed income instrument in US is price in USD, if the bank buy bond, after USD depreciate and the bank sell bond, it is still left with depreciated USD?

2) He mentioned he track CB. However, CB engagement in the market is not to make profit, their main purpose is to carry out monetary policy, or stabilize the market.

3)Currency Dealer do not carry inventory overnight, in fact the inventory turnover is fast.

4)he is using historical correlation, and select asset class based on his own weighting to create a synthetic. However there are a few problem,
i)How much deviation from prime asset and the basket asset does one has to see before entering while not suffering large DD?

ii)The correlation can break anytime based on the development in the economic front, how can the user spot it fast enough before getting fool by the value?

Everyone has some ideas please feel free to chip in, thanks

regards

Hi UnnamedPlayr

I can't see your photos uploaded in imageshack, can you reupload there photos in other sever ?

Many thanks !!!

5/9/2020, 3 year ...
After wandering through the darkness for a long time, I finally thought I had found something. The imbalance and rebalance that UP says might be like this:

I leave this as a compass for those of you who are still lost in there
Thanks UP for bringing us a gold topic !!!

It seems to be close to UP's point of view. I have studied for a long time. Can you tell me about the two kinds of imbalance? Forgive my bad English and hope to see your answer