Deferred Policy Acquisition Costs, Deferred Sales Inducements and Liability for Lifetime Income Benefit Riders

Policy acquisition costs deferred and amortized are as follows:



	December 31,
	2018			2017			2016
	(Dollars in thousands)
Balance at beginning of year	$	2,714,523		$	2,905,377		$	2,905,136
Costs deferred during the year:
Commissions	384,432			401,124			538,863
Policy issue costs	3,790			5,517			4,462
Amortization:
Amortization	(358,563		)	(304,162		)	(325,848		)
Impact of unlocking	30,572			48,198			(48,164		)
Effect of net unrealized gains/losses	761,084			(341,531		)	(169,072		)
Balance at end of year	$	3,535,838		$	2,714,523		$	2,905,377

Sales inducements deferred and amortized are as follows:



	December 31,
	2018			2017			2016
	(Dollars in thousands)
Balance at beginning of year	$	2,001,892		$	2,208,218		$	2,232,148
Costs deferred during the year	179,465			216,172			353,966
Amortization:
Amortization	(243,666		)	(210,886		)	(215,406		)
Impact of unlocking	21,465			34,274			(35,760		)
Effect of net unrealized gains/losses	557,565			(245,886		)	(126,730		)
Balance at end of year	$	2,516,721		$	2,001,892		$	2,208,218

The following table presents a rollforward of the liability for lifetime income benefit riders (net of coinsurance ceded):



	December 31,
	2018			2017		2016
	(Dollars in thousands)
Balance at beginning of year	$	704,441		$	533,391	$	351,946
Benefit expense accrual	157,333			149,442		139,443
Impact of unlocking	(53,607		)	21,608		42,002
Claim payments	—			—		—
Balance at end of year	$	808,167		$	704,441	$	533,391

We periodically revise the key assumptions used in the calculation of amortization of deferred policy acquisition costs and deferred sales inducements retrospectively through an unlocking process when estimates of current or future gross profits/margins (including the impact of realized investment gains and losses) to be realized from a group of products are revised. In addition, we periodically revise the assumptions used in determining the liability for lifetime income benefit riders as experience develops that is different from our assumptions.

The most significant revisions made during 2018 as a result of our quarterly reviews were account balance true-ups which were favorable to us due to stronger index credits than we assumed due to strong equity market performance and adjustments to generally decrease lapse rate assumptions to reflect better persistency experienced than assumed. The favorable impact of the account balance true-ups and lapse rate assumption changes was partially offset by revisions to lower our future investment spread assumptions primarily due to an increase in the cost of money we have been experiencing.

The most significant revisions made during 2017 as a result of our quarterly reviews were account balance true-ups which were favorable to us due to stronger index credits than we assumed due to strong equity market performance and adjustments to generally decrease lapse rate assumptions to reflect better persistency experienced than assumed. The favorable impact of the account balance true-ups and lapse rate assumption changes was partially offset by reductions in estimated future gross profits attributable to revisions to assumptions used in determining the liability for lifetime income benefit riders as well as an increase in estimated expenses associated with a reinsurance agreement with an unaffiliated reinsurer.

The most significant revisions during 2016 as a result of our quarterly reviews were adjustments to lower future spread assumptions as actual investment spreads being earned showed investment spread and gross profits being less than what we were assuming in our models due to decreases in the average yield on invested assets resulting from the continued low interest rate environment. We also made adjustments to extend the period of time in which we assume investment spread will grade up to our long-term spread targets by an additional two years as yields obtained on investment purchases were much lower than we had anticipated as a result of the overall decline in investment yields that followed the Brexit vote. In addition, revisions to assumptions used in determining the liability for lifetime income benefit riders during 2016 resulted in a decrease in estimated future gross profits.

The 2018, 2017 and 2016 revisions to the liability for lifetime income benefit riders were consistent with the revisions used in the calculation of amortization of deferred policy acquisition costs and deferred sales inducements described above. The 2018 revisions were primarily attributable to account balance true-ups and future investment spread assumptions. The impact of the account balance true-ups and future investment spread changes was partially offset by the lapse rate assumptions changes described above. The 2017 revisions were primarily due to the lapse rate assumption changes described above and changes to our account value growth projections. The 2016 revisions were primarily due to actual index credits on policies being lower than projected over the past four quarters.