Debate at our firm regarding the number of futures to use when constructing libor discount curve.One argument, which we have abided by historically, is to use the most liquid instrument at each maturity range along the curve i.e. for USD cash short end, short futures out to ~4yrs, swaps beyond.Another argument, being promoted by a new group of traders, is to use the full strip of futures (i.e. out to 10yrs or so - even though illiquid) to avoid problems with interpolation (zigzaggy forwards).I would have thought that the latter has several issues: illiquidity/stale data, convexity adjustments, replicating par swap rates etc.What is market practice?Any opinions either way much appreciated.Cheers,Jon

For this case, your firm's history is a good teacher. And typically cubic spline interpolation is needed to reduce the effect of the kink at the point where the futures meet the swaps. After all is done, always plot the implied forward curve to check if anything seems badly wrong.

Slightly confused by your question, you are asking about how to construct a discount curve, but the traders are worried about the forward curve.I suspect you are using a "one curve" methodology as opposed to a two curve (separate discount and forward curves) methodology.Focussing on your question (and a one curve approach) I suggest you look at the problem from two angles:a) Valuationb) Risk (specifically Market Risk)Ask the traders what the trade off is between smooth forwards versus valuation / risk derived from an illiquid market. I have seen occasions where people have used implied futures prices as opposed to market quotes to derive a "better" forward curve.Another approach to take is to see how flexible your yield curve bootstrapping algorithm is and whether it is possible to make changes there to create a more realistic forward curve. Ideally you want to use liquid market prices / rates and a theoretically consistent bootstrapping algorithm.CheersQuoteOriginally posted by: jonDebate at our firm regarding the number of futures to use when constructing libor discount curve.One argument, which we have abided by historically, is to use the most liquid instrument at each maturity range along the curve i.e. for USD cash short end, short futures out to ~4yrs, swaps beyond.Another argument, being promoted by a new group of traders, is to use the full strip of futures (i.e. out to 10yrs or so - even though illiquid) to avoid problems with interpolation (zigzaggy forwards).I would have thought that the latter has several issues: illiquidity/stale data, convexity adjustments, replicating par swap rates etc.What is market practice?Any opinions either way much appreciated.Cheers,Jon

I would go one step further and consider tension splines - this will improve the handling of the futures rollQuoteOriginally posted by: alvinkamFor this case, your firm's history is a good teacher. And typically cubic spline interpolation is needed to reduce the effect of the kink at the point where the futures meet the swaps. After all is done, always plot the implied forward curve to check if anything seems badly wrong.

QuoteOriginally posted by: jonDebate at our firm regarding the number of futures to use when constructing libor discount curve.One argument, which we have abided by historically, is to use the most liquid instrument at each maturity range along the curve i.e. for USD cash short end, short futures out to ~4yrs, swaps beyond.Another argument, being promoted by a new group of traders, is to use the full strip of futures (i.e. out to 10yrs or so - even though illiquid) to avoid problems with interpolation (zigzaggy forwards).I would have thought that the latter has several issues: illiquidity/stale data, convexity adjustments, replicating par swap rates etc.What is market practice?Any opinions either way much appreciated.Cheers,JonBoth ways of looking at your curve risk are useful, eg:1) liquid instrumentsYou know what the prices are, very handy. But your deltas are not on a consistent set of instruments. How are you going to visualise 3s/6s risk? fra/ois risk?2) extended futures stripYou get a lovely consistent use of end-to-end forwards, great for spreads, but you have to generate the futures prices out past the first few years yourselfSo how to get the best of both worlds?Get a decently granular set of input info, interpolate sensibly, and then transform risk against multiple potential sets of "input instruments".But most people do not do this very well, and it is an endless source of amusement to me that people put so much intellectual effort into sophisticated models of forward volatility while being picked off on their underlying forwards.You won't be all that good unless you can encompass both viewpoints on the curve, because they're both valid and useful. It's not an either-or thing. If you want to broaden your horizons further, talk to a good short rate trader. You are missing out at least one alternative way of looking at things in the front 2y of the curves.And perhaps Martinghoul will come in on this if he can be arsed

I am with ark on this, in terms of the general philosophy of these things. These are all complicated issues with no "either/or" correct answers, IMHO.However, one constant question, regardless of the setting, is the question of what it is that you're trying to achieve. If you're tinkering with the ultra-short stuff in FRA and OIS you will potentially want to concentrate your efforts in a completely different area than if you were looking at the pricing of some distant fwd swaps. So all you need to do is ask yourself a question of what sort of pricing is important to you. You'll then be able to make informed decisions regarding the cost/benefit of the various available options.Personally, I have always emphasized liquidity over consistency. As such, I only use a limited number of liquid front futures for building my curves. But that's just a specific choice I have made in my particular situation. If I had lots more time/resources, there's all sorts of wonderfully useful things that can be done, along the lines of what arkestra suggests.

The more futures contracts you use the more carefully you will need to consider a convexity correction.

One common idea as well is to fit to liquid instruments, derive the impliy futures prices from these, then refit your curve to these implied prices and compute risks w.r.t. to these. This is known as "resampling" of course.It helps you in two ways: 1) see how far off you interpolation is to illiquid prices (and since they are illiquid you may be within bid/offer spreads anyway) 2) express your risk in terms of instrument sthat may be more intuitive to your trader(s).But clearly as mentioned before, it all depend on what you want to achieve exactly. Resapling doesn't guaranee smooth forwards for instance.

instead of focusing on USD let's do a generic argument.I'd say it doesn't matter if you use futures or use swap. there is no liquid long term futures anyway but the point is if you use futures to build your curve you are constraining the forwards directly. this will help traders value either FRA or forward starting swaps more accurately.but of course whatever rate you input you must price back swaps to par, be it in a single curve or dual curve environment.The other constrain is that the projected libors must be reasonable; ie smooth and realistic.There is no problem with convexity cause the inputs are not exactly futures rate anyway. I'll support futures curve even for a emerging market curve.

Good Morning I am confused about the two curves methodology I wonder how you are computing you forward rate and more specifically which discount factor you are using. Are you still using the standard Forward rate formula: [Px(t,T1) - Px(t,T2)]/[(T1-T2)*Px(t,T2)] or are you discounted using the funding curves (such as OIS) that we called the "discount curves" and in that case we rather use : 1/(T1-T2)*[1/DiscCurve(t,T1,T2)]. ? For example i wonder how you calculate your Forward Rate for a FRA 3x6 ?Many thanks Best regards

You mean cubic interpolation reduce the kinks in fwd curve, right ? How about the hedges ? Usually the hedge seems to jump a lot when one new future rolls in. Anybody has experiences and solutions on this ? Greatly appreciated !