This is like weather forecast. The longer is interval - the lower is accuracy. I guess that ex post this IMF forecast till 2021 will be half wrong. Till 2035 I guess that there is no model even in IMF.

There are too many random factors, like global and local crises. The simplest thing is to extrapolate the average growth rate for the last let say 5 years into future. Those who grew on average with 5% per year are expected to continue with the same speed, and those who did with 1%  will continue with 1%. China had close to 10% growth for 2 decades, so the error ex post will be not high. But many countries fell into unexpected crises. The high volatility of oil prices had influenced the growth rate of many oil exporters, but not only them. Even the average global growth in the next year is often revised by serious institutions like IMF or WB.

Thanks Yuri. This is insightful

Dear Irabor,

As pointed by Yegorov, the longer the interval the lower the accuracy. This explains why static forecasting is usually more accurate compared to dynamic forecasting. 

However, you may go ahead and get the forecast estimates by using an Auto-regressive model (AR). I suggest you use Zaitun time series analysis software. I have herein attached a video as well as some guiding manuals on the same. 

I hope this helps. Thank you.

PPP method = population growth, participation rate, productivity (i.e. per worker real GDP growth)

PPP method is very simple and the lags can be minimized

.

I support Yuris point. Therefore I would not use the PPP method because it means to make a sort of weather forecast for 3 variables instead of one. Apart from the inevitable forecast errors in general, there is also the risk that the definition of GDP has changed and will be changed from time to time. It may even be that it will be of minor importance in 2035. GDP is an indicator for several things (production, income, welfare), which are, in general, related, but not the same. In this sense, it is an overloaded indicator.

If you cannot avoid a forecast, you should make it as simple as possible. I would take the same growth rate for all low-income countries and one (lower) rate for those with middle income. If the statistics for the past shows extremely high rates (which I think is unlikely), then one can assume that this rate can hold for the whole time and the rate should be lowered for the second decade.

It is almost imposible forecast long run till 2035, but make a scholastic exercise by using lagged dependent variablel econometric regression model: GDP = GDP(t-1) + bX + D + U.

Where:

X include x1: exportation, x2: importations, x3: capital flows, x4 employments, D: time dummies

Taking account at least 10 years data 

Then you make a simulation till 2035.

Mario's method is even more complicated than PPP.