Northcoders Share Discussion Threads

[zzaxx99] 23/9/2002 22:54

There may well be codes on the trades transmitted into the LSE (and in fact must be, thinking about it), but on the trades published by the LSE (which is what we see on the Trades list), I'm 99.99999% convinced that the broker / MM are not identified.

I believe that the published transaction code forms a unique key into a transactional record within the LSE, but that the details of the record to which that key relates, are not released outside the LSE. In fact, the LSE guarantees anonymity of the counterparties, so (although some educated guessing can take place), the code cannot have identity encoded within in.

Before that thread appeared on the PBB, I'd done a bit of peering at trade reports myself - my conclusion then was that, on stocks with few MMs there were too many different codes - on stocks with many MMs there were not enough codes - this was borne out by the discussion excerpted above.

[marquis] 23/9/2002 17:22

zzaxx99 - 23 Sep'02 - 13:09 - 5 of 8

May I add thanks to your time and effort.

I copy the page you suggested;

"With effect from Thursday 7 February 2002, Robert W Baird Limited ("Robert W Baird" or "the firm") will trade some of its agency and principal business through the following new additional set of trading codes:

Firm Code: 302
CREST Code: 601
SEAQ Code: GRAN
BIC Code: GRABGB21

Robert W Baird will continue to act in a principal and agency capacity and as a market maker in UK equities. The firm continues to have access to the SETS order book, the International Order Book and is a registered participant in the International Retail Service with the ability to enter orders.

The firm has a Model A settlement agreement with Pershing Limited and for its agency business, a Model B settlement agreement with Pershing Securities Limited.

Details are as follows with changes in bold:




Robert W Baird Limited
Mint House
77 Mansell Street
London
E1 8HF


Telephone:
020 7488 1212 (General)

Facsimile:
020 7481 3911



Additional Model B - agency/principal-SETS-SEAQ-IRS-SEATS-IOB

Firm Code:
302

CREST Code:
601

SEAQ Code:
GRAN

BIC Code:
GRABGB21


Model B - agency/principal-SETS-SEAQ-IRS-SEATS-IOB

Firm Code:
302

CREST Code:
601

SEAQ Code:
GRAV

BIC Code:
GRDAGB21


Model A - principal-Market Making

Firm Code:
322

CREST Code:
322

SEAQ Code:
GRAB

BIC Code:
GRBLGB21"

Do these codes(Firm, Crest, Seaq and BIC) not translate to identifying which stock broker is trading in the transaction code list? From your comments you suggest not.

In answer to my original question;
"Transaction codes on Stocks: Is there a Transaction Code Directory anywhere?"

I take it that you are suggesting there is not.

[zedder] 23/9/2002 14:15

zzaxx99,
Thanks for your time and effort. I'd been about to ask the question about codes recently when I discovered Marquis already had.
I'd been hoping that the first 3 alpha-numerics would have related to particular MMs, the theory being that you might be able to see if one was absorbing stock ready to fill a big buy order.

[zzaxx99] 23/9/2002 13:13

sorry - have another look at #5 - I've cleared up most of the formatting, and am just editing the last bits - should be a bit more readable now!

[zzaxx99] 23/9/2002 13:09

If you have access to the PBB, have a look at - which goes into this in extensive depth. If you don't have access, here are some relevant highlights:Here's some info - well lots actually, on individual Marker Makers and their codes, Crest, Firm Code, SEAQ, BIC Code. Doesn't seem to be a connection with any of those codes and the individual trade codes. In fact the trade code only needs to be a unique code that can be cross-referenced with another set of records containing all the necessary information - no need for it to be embedded....as everyone has said sadly the year number at the end is actually the most useful piece of information.The first digit or letter identifies a sequence of numbers. There appear to be 16 sequences from which numbers are allocated (perhaps there are 16 servers receiving trades from the market). The digits after the first digit (apart from the last two which are the year) form the sequence number, again using both letters and digits. If you take all the trades with codes beginning with a given digit or letter, you will see that the sequence number increments with each successive trade, taken across the whole market chronologically. Not very useful.There are often batches of trades in which the variable elements are minimal; only one character differing in each ten-character set.Eg: Go the top page of today's trades (25 June) in BT.A and look at AT trades numbered 2069-2072 timed at 16:27:30.The first character is from the usual 16; the letters A-F and numbers 0-9 (randomly selected server? but in this case not changing). And the last two as usual are the year 02.Only the last of the middle seven characters differs..Thus:2072.... 901XUK3Q022071.... 901XUK3O022070.... 901XUK3M022069.... 901XUK3K02Uniqueness attributable to only one alphabetic character? What does that tell us?;O)PS: If I was able to pick that out of the first page I looked at, I presume there must be millions of such examples out there.PPS: If these 'unique transaction identifiers' involve only 36 choices of character (26 letters and 10 numbers) in each of seven positions, I presume they are only used to distinguish between trades within a set period - and can be reused the next day, next week, next month, whatever. In which case does something within the code identify the time period in which it was used? (it might be a sequential rolling stream within one column - but the sequence need not be a standard numeric one). There aren't enough possible combinations to use for individual trades ad infinitum.... When they get to the end of the odd letters they then run through the odd numbers, then the even letters/numbers, and repeat. But the trades listings are full of such batches with only one character differing. Just going back one page on BT.A (and that's enough for me now thanks..) look at 2001-2003, 2042-2045 (where the sequence is 5,7,9,B) and 2059-2061. Always using alternate sequence. Giving no more than 36 differs before some other column needs to change. Here are a nice set of trade codes in sequence for TSCO today going up in odd twos.A01VAFHF02A01VAFHD02A01VAFHB02A01VAFH902A01VAFH702A01VAFH502 and another4011EYU5024011EYU3024011EYU1024011EYTZ02 showing Z going to 1 and the T clocking up to UThis looks pretty much like a sequence number and not a code. If you were thinking of designing a system to uniquely identify trades then I think you would come to the conclusion that the sequences should have no meaning of their own. They should only have a meaning when looked up on the LSE server. I can't believe that the designers of the computer system would have embeded anything in the code itself. There is no need to do this. If the number is mearly a look up reference then it makes the system more flexible. The data items being looked up via the sequence number can then be as rich in meaning as is needed and can me modified and added to later with the minimum of alteration. I think this is a dead end. How the different sequences are chosen is more of a problem. It may be that it changes after a certain number of trades on the LSE or it may be pseudo random. Re: when the numbers recycle - does anyone know how any trades there are on the LSE in a given period (day/month/year) - not the volume traded, but the number of discrete trades.If the number sequence uses only the 7 digits, then there are 36^7 combinations, or 78,364,164,096 different trades that can be uniquely identified before the codes repeat.Give that the last 2 digits represent the year, ISTM that this is likely to represet a year's worth of trades - but can anyone confirm that this volume is reasonable?Thinking about it, it should be possible to work this out. There are something like 3500 stocks; VOD does maybe 7000 trades per day. At most then, there's 3500 x 7000 trades per day (a wild exagerration, as most stocks won't do anyting like VOD's trading, but let's see how the numbers come out) - that's 24,500,000 individual trades per day. That allows for 3,198 days of 24.5m trades per day - comfortably 10 times more than the most wildly overstated number of trades likely in a year.Re: the alternate letters issue raised earlier - I don't think there is anything terrible significant about this - I suspect that there were a rapid spate of BT trades that were assigned nearly-contiguous numbers; but across the market aas a whole, there were trades happening at the same time that got assigned numbers in between. Pretty coincidental that that sequence would all step in 2s rather than by a random interval, but it was only a single sample of trades, and is not impossible.For example, if you look at these trades from VOD for today:Num.CodePriceSizeTypeCTBidOffer Time 5829F01Y18ET0286.53446AT 86.2586.5 16:20:103,446 242,250,534 224,014,888 27,383,808 5828D022D3C50286.2510381AT 86.2586.5 16:20:09 10,381 242,247,088 224,014,888 27,383,808 58275027FA5C0286.510206AT 86.2586.5 16:20:0610,206 242,247,088 224,004,507 27,383,808 5826102190GN0285.524624O L86.2586.5 16:01:20 4,624 242,236,882 224,004,507 27,383,808 58255027FA510286.525000AT 86.2586.5 16:19:5625,000 242,236,882 224,004,507 27,379,184 5824601ZIDE70286.527703AT 86.2586.5 16:19:4827,703 242,211,882 224,004,507 27,379,184If you look at trades 5825 & 5827 you'll see that the last (pre-year) digit increments from 1 to C - a jump of 11 – which isn’t terribly obvious significant!I don't think the use of alternate numbers is for the coincidental reason you suggest - it is consistent, even when rolling over from numbers to letters to numbers within a run.Anyway, between us, I think we've demolished the code idea. Pity. I like unravelling codesI have found a use for the sequences though. If you have a stock where a trade is in the undefined column rather than an obvious buy or sell then it is possible to see sometimes (and I stress sometimes) the real sequence and it's buy/sell status. You need to look at a couple of heavily traded stocks such as TSCO, VOD etc. Looking at a few of these will tell you the actual time sequence the trade was reported at. It doesn't always work, because the trade may have been notified late already. However I have used it to work out when some trades were done. It works particilarly well on trades done at the start of day and delayed. It works because you know it cannot have startd before a certain time in this case. I have found trades at 10:00 that were allocated a sequence number at around 8:00 and looked like a buy before I checked. This is interesting on lightly traded stocks.No significance to italicisation above, just used to identify different messages. Hope that clears it all up!

[crystalclear] 23/9/2002 13:09

What do you call a deer with no eyes?
No idea!

[marquis] 23/9/2002 11:27

I ask the same questions again today, as I have on the opening page.

Comments please

[marquis] 19/8/2002 08:49

And again!!

[marquis] 17/8/2002 10:07

And again!

[marquis] 16/8/2002 09:16

I ask the same questions again today, as I have on the opening page.

Comments please

[marquis] 15/8/2002 18:17

Transaction codes on Stocks: Is there a Transaction Code Directory anywhere?

The reason I ask this question?
I viewed the transactions today (15th Aug 2002) for Avon Rubber. There were 16 trades(including a zero trade) as you will see;

15 2027TYHP02 164.0 2000 O 161.0 164.0 14:14:00 2,000
35,805 16,500
14 000MRFF602 164.0 3000 O 161.0 164.0 13:47:00 3,000
33,805 16,500
13 1022CXHM02 161.0 500 O 161.0 164.0 13:31:00 500
30,805 16,500
12 1022CYX402 164.0 7800 O 161.0 164.0 13:14:07 7,800
30,805 16,000
11 2027U5CU02 161.0 6000 O 161.0 164.0 11:40:00 6,000
23,005 16,000
10 000MRHMU02 161.75 5000 O 161.0 164.0 11:25:43 5,000
23,005 10,000
9 1022CZKS02 162.0 5000 O 161.0 165.0 11:04:37 5,000
23,005 5,000
8 3029A7F402 165.0 2000 O 163.0 165.0 10:12:25 2,000
23,005
7 3029A46502 164.8 1455 O 163.0 165.0 9:54:00 1,455
21,005
6 000MRG0Y02 166.5 2376 O 163.0 167.0 8:14:00 2,376
19,550
5 2027U30Z02 165.0 5000 O 160.0 165.0 8:10:00 5,000
17,174
4 1022D02602 155.0 5000 M 153.0 155.0 8:05:48 5,000
12,174
3 1022CP0T02 154.65 5000 O 153.0 155.0 8:04:00 5,000
7,174
2 2027TY0R02 155.0 900 O 153.0 155.0 8:03:00 900
2,174
1 2027U70H02 154.65 960 O 153.0 155.0 8:01:00 960
1,274
0 1022CQ0V02 155.0 314 O 153.0 155.0 8:00:00 314
314

Observation reveals a big hike in purchase price from trade 4 (155) to trade 5 (165).
Then I notice trades 0, 3, 4 all start with the transaction code 1022. Is that the same stockbroker?

Transaction code 1022 also for trades 9, 12, 13.

If 1022 relates to a certain broker, could the market makers know that he has good knowledge of the market and so upgrade the shares.

Is this something 'we' as smaller players should be aware of?

Comments please.

[m.t.glass] 26/6/2002 18:18

Next question then: Is the rolling sequence used in columns 2-7 the same as the sequence that we now know (until they change it!) is used in column eight - namely ...5-7-9-A-C-E... ...6-8-0-B-D-F.. etc

Maybe the only way to tell (because the earlier columns don't tumble as fast) is to pick one of the most frequently traded stocks on one of the quietest days of the year. Something to do instead of the jumbo Xmas crossword :o)

[haystack] 26/6/2002 17:52

I have found a use for the sequences though. If you have a stock where a trade is in the undefined column rather than an obvious buy or sell then it is possible to see sometimes (and I stress sometimes) the real sequence and it's buy/sell status. You need to look at a couple of heavily traded stocks such as TSCO, VOD etc. Looking at a few of these will tell you the actual time sequence the trade was reported at. It doesn't always work, because the trade may have been notified late already. However I have used it to work out when some trades were done. It works particilarly well on trades done at the start of day and delayed. It works because you know it cannot have startd before a certain time in this case. I have found trades at 10:00 that were allocated a sequence number at around 8:00 and looked like a buy before I checked. This is interesting on lightly traded stocks.

[m.t.glass] 26/6/2002 16:52

zzaxx99 - I don't think the use of alternate numbers is for the coincidental reason you suggest - it is consistent, even when rolling over from numbers to letters to numbers within a run.

Anyway, between us, I think we've demolished the code idea. Pity. I like unravelling codes.

[m.t.glass] 26/6/2002 16:47

LSE should cope for a while yet then Haystack :o)

Pity BT never manage to build in such redundancy every time they introduce changes to phone numbers..

[zzaxx99] 26/6/2002 16:44

Re: when the numbers recycle - does anyone know how any trades there are on the LSE in a given period (day/month/year) - not the volume traded, but the number of discrete trades.If the number sequence uses only the 7 digits, then there are 36^7 combinations, or 78,364,164,096 different trades that can be uniquely identified before the codes repeat.Give that the last 2 digits represent the year, ISTM that this is likely to represet a year's worth of trades - but can anyone confirm that this volume is reasonable?Thinking about it, it should be possible to work this out. There are something like 3500 stocks; VOD does maybe 7000 trades per day. At most then, there's 3500 x 7000 trades per day (a wild exagerration, as most stocks won't do anyting like VOD's trading, but let's see how the numbers come out) - that's 24,500,000 individual trades per day. That allows for 3,198 days of 24.5m trades per day - comfortably 10 times more than the most wildly overstated number of trades likely in a year.Re: the alternate letters issue raised earlier - I don't think there is anything terrible significant about this - I suspect that there were a rapid spate of BT trades that were assigned nearly-contiguous numbers; but across the market aas a whole, there were trades happening at the same time that got assigned numbers in between. Pretty coincidental that that sequence would all step in 2s rather than by a random interval, but it was only a single sample of trades, and is not impossible.For example, if you look at these trades from VOD for today:Num.CodePriceSizeTypeCTBidOffer  Time  5829F01Y18ET0286.53446AT  ;86.2586.5 16:20:103,446  ;    242,250,534 224,014,888 27,383,808 5828D022D3C50286.2510381AT  86.2586.5 16:20:09 10,381     242,247,088 224,014,888 27,383,808 58275027FA5C0286.510206AT  86.2586.5 16:20:0610,206      242,247,088 224,004,507 27,383,808 5826102190GN0285.524624O L86.2586.5 16:01:20  4,624    242,236,882 224,004,507 27,383,808 58255027FA510286.525000AT  86.2586.5 16:19:5625,000     ; 242,236,882 224,004,507 ;27,379,184 5824601ZIDE70286.527703AT  86.2586.5 16:19:4827,703   ;   242,211,882 224,004,507 27,379,184If you look at trades 5825 & 5827 you'll see that the last (pre-year) digit increments from 1 to C - a jump of 11 – which isn’t terribly obvious significant!

[haystack] 26/6/2002 16:30

The sequence seems to be made up of 8 characters. The one at the front has 16 possible values. The others have 36 values. That's (36^7) times 16. I make that 1,253,826,625,536. If only every other number is used then that comes to 626,913,312,768 trades. I wouldn't think that you would run out of numbers in a year. That's around 2,500,000,000 per day. That's trades and not numbers of shares.

[m.t.glass] 26/6/2002 14:49

Automatically rerunning after a complete spin of all available numbers/letters would reduce the number of duplicated IDs, and would ensure the system did not fall foul of exceptional volumes within any fixed timespan I suppose. And presumably any enquirer seeking a particular trade is soon going to spot, via other criteria, which of several with the same ID is the one that fits the search.

[haystack] 25/6/2002 21:40

Here are a nice set of trade codes in sequence for TSCO today going up in odd twos.
A01VAFHF02
A01VAFHD02
A01VAFHB02
A01VAFH902
A01VAFH702
A01VAFH502

and another

4011EYU502
4011EYU302
4011EYU102
4011EYTZ02

showing Z going to 1 and the T clocking up to U

This looks pretty much like a sequence number and not a code. If you were thinking of designing a system to uniquely identify trades then I think you would come to the conclusion that the sequences should have no meaning of their own. They should only have a meaning when looked up on the LSE server. I can't believe that the designers of the computer system would have embeded anything in the code itself. There is no need to do this. If the number is mearly a look up reference then it makes the system more flexible. The data items being looked up via the sequence number can then be as rich in meaning as is needed and can me modified and added to later with the minimum of alteration. I think this is a dead end. How the different sequences are chosen is more of a problem. It may be that it changes after a certain number of trades on the LSE or it may be pseudo random.

[kayak] 25/6/2002 20:17

Think in terms of a base 36 sequence number... e.g. after 0000009 would come 000000A through 000000Z and then 0000010 etc. Well spotted about the odd/even business though.

[m.t.glass] 25/6/2002 19:26

The odds/evens thing applies to the 36 differs used in the eighth column Baz. The 16 differs A-F and 0-9 are in column one, apparently in random order.

Going now.

I am not an anorak
I am not an anorak
I am not an anorak

[bristolbaz] 25/6/2002 19:21

don't forget that A=10 and F=15

[m.t.glass] 25/6/2002 19:20

I'll go with zzaxx99:

"..I'm 100% convinced that this is a centrally-assigned unique identifier that has no useful meaning to anyone apart from the LSE's software..."

[jonc] 25/6/2002 19:11

MT,

I suppose anything is possible.

But as Bully said earlier if Paul phones the LSE they will tell him there is no significance in the codes for us mere mortals.

Cheers

JC

[m.t.glass] 25/6/2002 19:08

No it's not that. When they get to the end of the odd letters they then run through the odd numbers, then the even letters/numbers, and repeat.

But the trades listings are full of such batches with only one character differing. Just going back one page on BT.A (and that's enough for me now thanks..) look at 2001-2003, 2042-2045 (where the sequence is 5,7,9,B) and 2059-2061. Always using alternate sequence. Giving no more than 36 differs before some other column needs to change.