The Link You Do Not Own

Two exposures, and they are not the same size

An outage can cost you two different things and almost every treatment of this subject mixes them. It can stop you sending an order, and it can leave a position on with nobody watching it. Those are separate exposures, they have separate fixes, and on lesson 71’s four rules they differ by three orders of magnitude.

Measure them on the same twenty-nine moves. The book asks for an order on thirteen of the twenty-nine days, and each of those asks is a single instant, because every rule in the grid reads one number at the close and nothing else. Call that instant a minute and the decision exposure is thirteen minutes. The position exposure is the other object: at least one of the four rules is holding on all twenty-nine of the days, around the clock, which is 41,760 minutes. The ratio is 3,212 to one. Be five times more generous about the instant and it is still 642 to one.

Everything below follows from that. Redundancy on your own side of the chain protects the smaller exposure. Only an order already resting somewhere else protects the larger one.

Start with the smaller exposure anyway, because the arithmetic people actually do is about the second connection. Take one link that is down one per cent of the time, which is 87.6 hours a year. Add a second and the textbook answer is that the pair is down 0.01 per cent of the time, or 0.9 hours, because 0.01 times 0.01 is 0.0001. That answer requires the two to fail independently, and the honest parameter is the share of failures common to both.

The first row is the promise and the fourth is the purchase. At one failure in five being common to both links, the pair is down 18.2 hours a year rather than 0.9, which is twenty times the promise, and nothing about the second link changed. This is lesson 71’s finding wearing different clothes: two things are two things only to the extent that they fail separately, and the share that does not is the number nobody measures. Two providers into one building share the building, the power and the street.

Now the larger point, which is that your side is not the whole chain. An order reaches the market through your equipment and then through the broker, and availability along a chain multiplies. So the broker’s figure is a ceiling on yours, and you cannot buy past it.

Read the third column down. The first purchase, from 99.0 to 99.9, is worth 78.8 hours a year. The second, from 99.9 to 99.99, is worth 7.9. The third is worth 0.8, and no purchase after that is worth anything at all, because the 8.8 hours that are left belong to the broker. Every treatment of this subject that lists equipment in descending price order has the sequence backwards: the money buys a lot once, a little twice, and then it is buying nothing at all.

The fourth column is the same story counted in the unit that matters. Lesson 76’s thirteen orders in twenty-nine days is 113 orders a year. At 99 per cent on your own side you miss 1.13 of them; at 99.99 per cent you miss one every eighty-nine years.

Almost nobody has multiplied their own availability by their broker’s and looked at the product. It is one multiplication, the inputs are a number you can measure over a month and a number the broker either publishes or does not, and the product tells you which of the two links your next purchase is actually aimed at.

Module 10’s list of what the day contains that is not the decision gains its second item: the chain, and which link on it your money reaches.

Eighty-seven hours for nothing

Take a working setup at the top of that table: your own side up 99 per cent of the time, a broker at 99.9, and lesson 71’s four rules holding a position on every one of the twenty-nine days. There are two things you can do next and only one of them has a price.

Leave a stop resting at the broker. The position is then protected by the broker alone, because a submitted order is executed by the machine holding it and does not need yours. Unprotected time falls from 96.3 hours a year to 8.8. Cost: nothing.

Buy a second connection instead, with one failure in five common to both. Your side improves from 99.0 per cent to 99.792, so the chain reaches 99.692 and unprotected time falls from 96.3 hours a year to 27.0. Cost: a bill every month.

The free move is worth 87.5 hours a year and the paid one is worth 69.3. The free move is larger, and it is larger before the bill is subtracted rather than after.

Then do both, in that order, and read what the second purchase actually bought. With the stop already resting, the position is protected at the broker’s 99.9 per cent whatever your own equipment does, so the second connection contributes exactly zero hours to the position. Its entire remaining benefit is the other exposure, the thirteen instants: it takes missed orders from 1.13 a year to 0.235, which is nine tenths of one order.

So the thing your backup connection protects, once the stop is resting, is nine tenths of one trade a year. Lesson 67 needed 589 trades to establish whether a tenth of an R was real, and this book produces 113 orders a year, so the trade the second connection saves you is one whose contribution to your result is smaller than anything you could measure in a working life. It is still worth having. It is not worth having first, and it is not worth what the shopping lists charge for it.

So put a stop at the broker on every open position tonight, and buy the second connection afterwards if at all.

What this does not settle

That the decision takes a minute. That is stated, not measured, and it is the weakest input on the page. It is also the input that cannot change the conclusion: at five minutes a decision the two exposures still differ by 642 to one, and at an hour each the position exposure is still larger. What would change the conclusion is a rule that decides continuously rather than at the close, and this course does not carry one.

That 99.9 per cent is your broker. It is a parameter, and no figure on this page is a measurement of any broker’s uptime, because this course has not measured one and will not invent one. Substitute your own and the tables move; what does not move is the shape, which is that the chain is a product and its ceiling is the link you do not own.

That a resting stop is a solved stop. It is not. It removes an availability problem and leaves lesson 21’s untouched: a resting stop is filled at the price available when it triggers, which in a gap is far from where you put it, and the slippage on that fill is the same whether you were watching or not. What the resting order buys is that the order exists at all when you cannot reach the market.

That the common-mode share is knowable. The first table turns entirely on a number almost nobody can produce for their own setup, because it is the share of your outages whose cause reaches both links, and you learn it from the outages you have already had. If you have had two, you do not know it. The safe reading of that table is the fourth row rather than the first.

That uptime is the failure that matters. This page priced availability, which is whether a thing answers. It said nothing about correctness, and a feed that is up and stale is worse than a feed that is down, because a feed that is down announces itself and a stale one is answered by lesson 76’s twelve operations without complaint. Every arithmetic on this page assumes the numbers arriving are the right numbers.

And the concession that costs most: this page priced the machinery in hours and said nothing about what any of it costs in money. Neither has the rest of the course. Every figure since lesson 63 has been a gross figure, before the largest recurring deduction a trading result meets, and lesson 78 is what that omission has been worth.